Apparatus for generating Artificial Intelligence training dataset at the interface driver level and method thereof

Abstract

The present invention discloses an apparatus and method for generating a dataset for artificial intelligence learning at the interface driver stage. Specifically, the present invention standardizes physical address-based data collected in a heterogeneous PLC environment into a multi-layered structure and integrates the physical address-based data standardized into the multi-layered structure with the state variable identifier system of the SECS / GEM standard to directly generate a dataset for artificial intelligence learning at the interface driver stage. This allows for the generation of a semantic-based AI learning structure at the data generation point (Driver layer) without undergoing a separate data lake and / or ETL (Extract-Transform-Load) reprocessing stage after the collection of physical address-based data.

Description

Technology Field

[0001] The present invention relates to an apparatus and method for generating a dataset for artificial intelligence learning at the interface driver stage, and more specifically, to an apparatus and method for directly generating a dataset for artificial intelligence learning at the interface driver stage by standardizing physical address-based data collected in a heterogeneous PLC environment into a multi-layered structure and integrating the physical address-based data standardized into the multi-layered structure with the state variable identifier system of the SECS / GEM standard. Background Technology

[0002] In smart manufacturing and dark factory environments, real-time collection, analysis, and autonomous control of equipment data serve as key elements, and the industrial equipment environment consists of a PLC-based control layer and a SECS / GEM-based equipment communication layer.

[0003] A PLC (Programmable Logic Controller) operates based on physical addresses in bit or word units, and the address itself does not contain process meaning.

[0004] On the other hand, in a SECS / GEM environment, facility status is managed as a semantic-based identifier through SVID (Status Variable ID).

[0005] Conventional technology adopts a structure in which PLC data and SVID data are stored independently and reconstructed on a separate server for AI learning.

[0006] This structure results in issues with post-definition of data semantics, incomplete event-unit synchronization, address disconnection problems when PLC programs are changed, inconsistencies between SVID and PLC, increased costs for reconstructing AI training data, and the absence of a real-time AI feedback structure.

[0007] Furthermore, while the implementation of a dark factory requires a prerequisite structure where data inevitably accumulates, conventional methods are merely lagging models that form an analysis structure only after data has been loaded. Prior art literature

[0008] Korean Registered Patent No. 10-2769340 [Title: Multi-channel communication control device and multi-channel communication control method for equipment automation] The problem to be solved

[0009] The objective of the present invention is to provide an apparatus and a method for directly generating a dataset for artificial intelligence learning at the interface driver stage by standardizing physical address-based data collected in a heterogeneous PLC environment into a multi-layered structure and integrating the physical address-based data standardized into the multi-layered structure with the state variable identifier scheme of the SECS / GEM standard.

[0010] Another objective of the present invention is to provide an apparatus and a method for generating a dataset for artificial intelligence training that pre-normalizes physical address-based data into a multi-layer standardization structure.

[0011] Another objective of the present invention is to provide an apparatus and a method for generating an artificial intelligence learning dataset that integrates logical tags with the SVID scheme of the SECS / GEM standard.

[0012] Another objective of the present invention is to provide an apparatus and a method for generating an artificial intelligence learning dataset that generates the dataset directly at the interface driver stage.

[0013] Another objective of the present invention is to provide an apparatus and a method for generating an artificial intelligence learning dataset that provides a structure for maintaining data continuity even when a PLC program is changed.

[0014] Another objective of the present invention is to provide an apparatus and a method for generating a dataset for artificial intelligence training that secures high-quality training data through an event-based synchronization and hybrid sampling structure. means of solving the problem

[0015] An apparatus for generating a dataset for artificial intelligence learning according to an embodiment of the present invention may include: a communication unit that collects physical address-based data from a PLC within an equipment; a PLC MAP standardization unit that converts the collected physical address-based data into a PLC Map of a preset multi-layered structure to define a logical tag; an SVID standardization unit that maps the logical tag to a Status Variable ID (SVID) of the SECS / GEM (Semiconductor Equipment Communication Standard / Generic Equipment Model) standard; a semantic group definition unit that defines a semantic group based on the defined logical tag or the mapped Status Variable ID (SVID); and a dataset schema definition unit that defines the structure of the dataset for artificial intelligence learning to be generated.

[0016] As an example related to the present invention, the above-mentioned preset multi-layer structure may include a physical address layer, a logical tag layer, an SVID standardization layer, a semantic group layer, and an artificial intelligence feature definition layer.

[0017] As an example related to the present invention, when a project file is uploaded according to user input, an integrated control unit may be further included that is configured to maintain the logical tag and schema structure even if the physical address related to the PLC is changed by linking with the PLC MAP standardization unit and the dataset schema definition unit.

[0018] A method for generating an artificial intelligence learning dataset according to an embodiment of the present invention may include: a step of collecting physical address-based data from a PLC within a facility by a communication unit; a step of defining a logical tag by converting the collected physical address-based data into a PLC Map of a pre-set multi-layered structure by a PLC MAP standardization unit; a step of mapping the logical tag to a Status Variable ID (SVID) of the SECS / GEM standard by an SVID standardization unit; a step of defining a semantic group based on the defined logical tag or the mapped Status Variable ID (SVID) by a semantic group definition unit; and a step of defining the structure of the artificial intelligence learning dataset to be generated by a dataset schema definition unit.

[0019] As an example related to the present invention, the artificial intelligence learning dataset to be generated is defined as a table structure and converted into a tensor structure or a vector structure according to the learning model input format, and may include a timestamp, equipment ID, lot ID, process step, SVID group, derived feature value, logical tag value, label value, version ID, raw value, and derived variable.

[0020] As an example related to the present invention, a dataset generation type interface driver, according to a user request, generates a data snapshot of one or more data related to the user request in units of semantic groups among a plurality of physical address-based data stored in a storage unit, according to the structure of the AI ​​learning dataset to be generated defined above, and converts it into an AI learning dataset structure; a data integrity verification unit performs a pre-set numerical validity verification, a state transition-based verification, a cross-tag consistency verification, and a process rule and physical constraint-based verification on the generated AI learning dataset to calculate an integrity score; a key tag condition evaluation unit checks whether a record is generated for the generated AI learning dataset based on a record generation condition for a key tag defined in the dataset schema definition unit; and a threshold comparison unit determines whether the calculated integrity score is greater than or equal to a pre-set threshold and whether the generated AI learning dataset satisfies the record generation condition, respectively. The method may further include the step of requesting record generation from a record generation unit by the threshold comparison unit when, based on the above judgment result, the calculated integrity score is greater than or equal to a preset threshold and the generated artificial intelligence learning dataset satisfies the record generation condition; and the step of storing the artificial intelligence learning dataset in the storage unit by the record generation unit in accordance with the record generation request from the threshold comparison unit.

[0021] As an example related to the present invention, the method may further include the steps of: selecting one artificial intelligence learning dataset from a list corresponding to artificial intelligence learning datasets stored in the storage unit according to user selection, and when a dataset format is selected, converting the selected artificial intelligence learning dataset into the selected dataset format by a serialization unit; transmitting the artificial intelligence learning dataset converted into the dataset format by an integrated control unit to an artificial intelligence server through the communication unit; receiving feedback corresponding to the artificial intelligence learning dataset converted into the corresponding dataset format transmitted from the artificial intelligence server in response to the transmission by the integrated control unit through the communication unit; and performing an SVID update function and a control function for the facility by the integrated control unit based on the received feedback. Effects of the invention

[0022] The present invention has the effect of enabling the creation of a semantic-based AI learning structure at the data generation point (Driver layer) without going through a separate data lake and / or ETL (Extract-Transform-Load) reprocessing step after collecting physical address-based data, by standardizing physical address-based data collected in a heterogeneous PLC environment into a multi-layered structure and integrating the physical address-based data standardized into a multi-layered structure with the state variable identifier system of the SECS / GEM standard to directly generate a dataset for artificial intelligence learning at the interface driver stage.

[0023] In addition, the present invention has the effect of forming a semantic-based logical tag system by pre-normalizing physical address-based data into a multi-layered standardization structure.

[0024] In addition, the present invention has the effect of enabling the construction of a PLC-SECS-AI integrated data model by integrating logic tags with the SVID system of the SECS / GEM standard.

[0025] In addition, the present invention has the effect of eliminating post-processing steps by directly generating a dataset for artificial intelligence learning at the interface driver stage.

[0026] In addition, the present invention provides a structure that maintains data continuity even when changing PLC programs, thereby having the effect of ensuring data continuity.

[0027] In addition, the present invention has the effect of optimizing training data density by securing high-quality training data through event-based synchronization and a hybrid sampling structure. Brief explanation of the drawing

[0028] FIG. 1 is a block diagram showing the configuration of a device that generates a dataset for artificial intelligence learning in the interface driver step according to an embodiment of the present invention. FIGS. 2 and 3 are flowcharts illustrating a method for generating an artificial intelligence learning dataset in the interface driver step according to an embodiment of the present invention. Specific details for implementing the invention

[0029] It should be noted that the technical terms used in this invention are used merely to describe specific embodiments and are not intended to limit the invention. Furthermore, unless specifically defined otherwise in this invention, the technical terms used in this invention should be interpreted in the sense generally understood by those skilled in the art to which this invention pertains, and should not be interpreted in an overly broad or overly narrow sense. Additionally, if a technical term used in this invention is an incorrect technical term that fails to accurately express the concept of the invention, it should be replaced with a technical term that can be correctly understood by those skilled in the art. Moreover, general terms used in this invention should be interpreted according to their prior definitions or the context, and should not be interpreted in an overly narrow sense.

[0030] Furthermore, singular expressions used in the present invention include plural expressions unless the context clearly indicates otherwise. Terms such as "composed of" or "comprising" in the present invention should not be interpreted as necessarily including all of the various components or steps described in the invention, and should be interpreted as meaning that some of the components or steps may not be included, or that additional components or steps may be included.

[0031] Additionally, terms including ordinal numbers, such as first, second, etc., used in the present invention may be used to describe components, but the components should not be limited by the terms. The terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component.

[0032] Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings. Identical or similar components are given the same reference number regardless of the drawing symbols, and redundant descriptions thereof will be omitted.

[0033] Furthermore, in describing the present invention, detailed descriptions of related prior art are omitted if it is determined that such descriptions could obscure the essence of the invention. Additionally, it should be noted that the attached drawings are intended only to facilitate an understanding of the concept of the present invention and should not be interpreted as limiting the concept of the present invention.

[0034] The interface driver step described in the present invention represents a software layer between an operating system (OS) and a hardware device in a system.

[0035] In other words, the interface driver stage described above is a stage where the operating system does not directly control the hardware, but instead transmits commands and exchanges data through the driver.

[0036] In other words, the entire process is represented as follows.

[0037] 1. Application Phase

[0038] - The user runs the program.

[0039] Example: A word processor requesting output to a printer.

[0040] 2. Operating System Stage

[0041] - The operating system receives and processes requests via system calls.

[0042] Examples: file system, memory management, device management, etc.

[0043] 3. Interface driver stage

[0044] - Acts as an intermediary bridge between the operating system and the device.

[0045] - Converts general operating system commands into hardware commands that the device can understand.

[0046] - Responsible for checking device status, data transmission, error handling, etc.

[0047] 4. Hardware stage

[0048] The actual device executes the command.

[0049] Examples: printer output, writing data to disk, network transmission and reception.

[0050] As such, the interface driver step described in the present invention may be a step located between the operating system step and the hardware step, and may transmit commands / data through the driver.

[0051] FIG. 1 is a block diagram showing the configuration of a device (10) that generates an artificial intelligence learning dataset in the interface driver step according to an embodiment of the present invention.

[0052] As illustrated in FIG. 1, the device (10) for generating an artificial intelligence learning dataset at the interface driver stage is composed of a communication unit (100), a storage unit (200), a display unit (300), a voice output unit (400), a PLC MAP standardization unit (500), an SVID standardization unit (600), a semantic group definition unit (700), a dataset schema definition unit (800), a dataset generation type interface driver (900), a data integrity verification unit (1000), a key tag condition evaluation unit (1100), a threshold comparison unit (1200), a record generation unit (1300), a serialization unit (1400), and an integrated control unit (1500). Not all components of the device (10) for generating an artificial intelligence learning dataset in the interface driver stage shown in FIG. 1 are essential components, and the device (10) for generating an artificial intelligence learning dataset in the interface driver stage may be implemented with more components than those shown in FIG. 1, or with fewer components.

[0053] The device (10) that generates the AI ​​learning dataset in the interface driver stage may be implemented in the form of a data center, web server, database server, proxy server, etc. Additionally, the device (10) that generates the AI ​​learning dataset in the interface driver stage may be equipped with one or more of a network load balancing mechanism or various software that enables the device (10) that generates the AI ​​learning dataset in the interface driver stage to operate on the Internet or another network, thereby enabling it to be implemented as a computerized system. Furthermore, the network may be an HTTP network, a private line, an intranet, or any other network. Moreover, the connection between a terminal (not shown) possessed by any user and the device (10) that generates the AI ​​learning dataset in the interface driver stage may be connected via a secure network so that the data is not attacked by any hacker or other third party. Additionally, the device (10) that generates the AI ​​learning dataset in the interface driver stage may include a plurality of database servers, and the database servers may be implemented in such a way that they are connected separately from the device (10) that generates the AI ​​learning dataset in the interface driver stage through any type of network connection, including a distributed database server architecture.

[0054] In addition, the device (10) that generates the artificial intelligence learning dataset in the above interface driver step operates in an edge device, cloud environment, etc.

[0055] The communication unit (100) communicates with any internal component or any at least one external terminal through a wired / wireless communication network. At this time, the external terminal may include a server (not shown), etc. Here, wireless internet technologies include Wireless LAN (WLAN), DLNA (Digital Living Network Alliance), Wibro (Wireless Broadband), Wimax (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), IEEE 802.16, Long Term Evolution (LTE), LTE-A (Long Term Evolution-Advanced), LTE-M (LTE-based maritime wireless communication), Wireless Mobile Broadband Service (WMBS), 5G network / 5G communication network, 6G network / 6G communication network, Wi-SUN (Wireless Smart Utility Network), NarrowBand-Internet of Things (NB-IoT), etc., and the communication unit (100) includes at least one wireless internet technology within the scope including internet technologies not listed above. Data is transmitted and received accordingly.In addition, short-range communication technologies may include Bluetooth, Bluetooth Low Energy (BLE), ANT, ANT+, Long Range (LoRa), RFID (Radio Frequency Identification), Infrared Data Association (IrDA), Ultra Wideband (UWB), ZigBee, Near Field Communication (NFC), Ultra Sound Communication (USC), Visible Light Communication (VLC), Wi-Fi, Wi-Fi Direct, Magnetic Secure Transmission (MST), Beacon, EnOcean, Near Field Magnetic Induction (NFMI), Z-WAVE, and SIGFOX. Furthermore, wired communication technologies may include Power Line Communication (PLC), USB, Ethernet, and Serial It may include serial communication, optical / coaxial cables, etc.

[0056] In addition, the communication unit (100) can mutually transmit information with any terminal via a Universal Serial Bus (USB).

[0057] In addition, the communication unit (100) transmits and receives wireless signals to and from a base station, the server, etc., on a mobile communication network built according to technical standards or communication methods for mobile communication (e.g., GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), CDMA2000 (Code Division Multi Access 2000), EV-DO (Enhanced Voice-Data Optimized or Enhanced Voice-Data Only), WCDMA (Wideband CDMA), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G network / 5G communication network, 6G network / 6G communication network, etc.).

[0058] Additionally, the communication unit (100) collects (or receives) one or more physical address-based data from a PLC (not shown) within the facility (not shown) under the control of the integrated control unit (1500). Here, the physical address-based data is the entire data managed by the corresponding PLC, and is a value stored in the PLC memory area (e.g., Bit, Word, Double Word, etc.), and includes variables used for facility control and status representation, varies depending on the type of facility and the PLC program design, and includes process data such as temperature, pressure, Real Time values ​​of facility utility controllers (e.g., Flow Rate, RPM, Voltage, Current, Cylinder Position Value, etc.), valve opening / closing status (e.g., On / Off Bit), process step code, facility operation status code (e.g., Run / Idle / Alarm), alarm flag bit, timer value, counter value, Recipe ID, Lot ID (e.g., when managed by the PLC).

[0059] At this time, the communication unit (100) can collect physical address-based data from each heterogeneous PLC environment.

[0060] In this way, the communication unit (100) can collect one or more physical address-based data in relation to the PLC.

[0061] The above storage unit (200) stores various user interfaces (UI), graphic user interfaces (GUI), etc.

[0062] In addition, the storage unit (200) stores data and programs, etc., necessary for the device (10) that generates an artificial intelligence learning dataset in the interface driver stage to operate.

[0063] That is, the storage unit (200) can store a plurality of applications (application programs or applications) running on the device (10) that generates an artificial intelligence learning dataset in the interface driver stage, data for the operation of the device (10) that generates an artificial intelligence learning dataset in the interface driver stage, and commands. At least some of these applications may be downloaded from an external server via wireless communication. In addition, at least some of these applications may exist on the device (10) that generates an artificial intelligence learning dataset in the interface driver stage from the time of shipment for the basic functions of the device (10) that generates an artificial intelligence learning dataset in the interface driver stage. Meanwhile, the applications are stored in the storage unit (200) and installed on the device (10) that generates an artificial intelligence learning dataset in the interface driver stage, and can be driven by the integrated control unit (1500) to perform the operation (or function) of the device (10) that generates an artificial intelligence learning dataset in the interface driver stage.

[0064] Additionally, the storage unit (200) may include at least one storage medium among Flash Memory Type, Hard Disk Type, Multimedia Card Micro Type, Card Type Memory (e.g., SD Memory, XD Memory, CF (compact flash) Memory, etc.), Stick Type Memory Stick, Magnetic Memory, Magnetic Disk, Optical Disk, RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), OTPROM (one-time programmable ROM), Mask ROM, and Flash ROM. Additionally, the device (10) that generates a dataset for artificial intelligence learning at the interface driver stage may operate a web storage that performs the storage function of the storage unit (200) on the internet, or may operate in relation to said web storage.

[0065] In addition, the storage unit (200) supports relational databases, time-series databases, NoSQL databases, in-memory databases, edge storage structures, etc. At this time, row address storage is optional and may not be a required element.

[0066] Additionally, the storage unit (200) stores one or more physical address-based data, etc., under the control of the integrated control unit (1500).

[0067] The above display unit (or display unit) (300) can display various content, such as various menu screens, using a user interface and / or a graphic user interface stored in the storage unit (200) under the control of the integrated control unit (1500). Here, the content displayed on the display unit (300) includes various text or image data (including various information data) and menu screens, etc., including data such as icons, list menus, and combo boxes. Additionally, the above display unit (300) may be a touch screen.

[0068] Additionally, the display unit (300) may include at least one of a Liquid Crystal Display (LCD), a Thin Film Transistor-Liquid Crystal Display (TFT LCD), an Organic Light-Emitting Diode (OLED), a Flexible Display, a 3D Display, an e-ink Display, a Light Emitting Diode (LED), a beam projector, a goggle-type VR, a hologram, and a Head-Up Display (HUD). Here, when the display unit (300) is implemented as a HUD, the display unit (300) may be equipped with a projection module to output information through an image projected onto a windshield or window.

[0069] In addition, the display unit (300) may be implemented as a touch screen by forming a layered structure with a touch input unit (not shown) or by being formed as an integral unit.

[0070] Additionally, the display unit (300) may include a transparent display. The transparent display may be attached to a windshield or a window.

[0071] In addition, the transparent display can display a predetermined screen while having a predetermined transparency. To have transparency, the transparent display may include at least one of a transparent TFEL (Thin Film Electroluminescent), a transparent OLED (Organic Light-Emitting Diode), a transparent LCD (Liquid Crystal Display), a transparent transparent display, and a transparent LED (Light Emitting Diode) display. The transparency of the transparent display can be adjusted.

[0072] Additionally, the display unit (300) displays one or more physical address-based data, etc., under the control of the integrated control unit (1500).

[0073] The voice output unit (400) outputs voice information included in a signal processed by the integrated control unit (1500). Here, the voice output unit (400) may include a receiver, a speaker, a buzzer, etc.

[0074] In addition, the voice output unit (400) outputs guidance voice generated by the integrated control unit (1500).

[0075] Additionally, the voice output unit (400) outputs voice information (or sound effects) corresponding to one or more physical address-based data, etc., under the control of the integrated control unit (1500).

[0076] The above PLC MAP standardization unit (or PLC MAP standardization engine) (500) converts the collected physical address-based data into a PLC Map of a preset multi-layer structure and defines (or sets) a Logical Tag. Here, the preset multi-layer structure includes a physical address layer, a Logical Tag layer, an SVID standardization layer, a semantic group layer, an artificial intelligence feature definition layer, etc. At this time, the Physical Address Layer preserves the collected physical address-based data as is and includes a configuration in units of bit, word, and double word. In addition, the Logical Tag Layer converts the physical address into a process semantic-based tag name and expands it into multiple logical tags according to process conditions even for the same physical address. In addition, the SVID Standardization Layer maps the logical tag to an SVID in order to integrate PLC and SECS data into the same identification system. In this case, the mapping (or SVID mapping) includes 1:1 mapping, 1:N mapping, N:1 aggregation mapping that aggregates multiple logical tags into a single SVID, conditional mapping, etc. Additionally, the Semantic Group Layer organizes semantic groups into units such as equipment status, process step, event, recipe, and Lot (or Work order). Furthermore, the AI ​​Feature Definition Layer includes AI feature definitions that include physical address values ​​(or raw values / physical address-based data), derived variables, time interval statistics, event-based labels, normal / abnormal classification tags, etc. Here, the derived variables include moving averages, change amounts, standard deviations, derivatives, etc.

[0077] In addition, the above-mentioned logical tag is a meta-information structure that assigns meaning to a physical address, and includes a logical tag ID (e.g., CH1_T01), a tag name (or semantic-based name) (e.g., CHAMBER_TEMP), a corresponding physical address (e.g., DW18001), a data type (e.g., U4), a unit (e.g., C), affiliated facility information, namespace information (e.g., including Factory, Line, Equipment / Process Step, etc.) (e.g., FAB1.L1.EQ001.CHAMBER_TEMP), creation version information (e.g., V1), and status / attribute information (e.g., whether it is Read Only, etc.).

[0078] In this way, the PLC MAP standardization unit (500) can define one or more logical tags by pre-normalizing one or more physical address-based data into a multi-layer standardization structure.

[0079] The above SVID standardization unit (or SVID standardization layer) (600) maps the logical tag to a status variable identifier (SVID: Status Variable ID) of the SECS / GEM (Semiconductor Equipment Communication Standard / Generic Equipment Model) standard.

[0080] That is, the SVID standardization unit (600) identifies a state variable identifier corresponding to a logical tag among the dictionary (or SVID by logical tag) for SVIDs stored in advance in the storage unit (200), and maps the identified state variable identifier corresponding to the logical tag to the logical tag. Here, the mapping includes conditional rule-based mapping, mapping that changes dynamically according to a recipe or process state, version control table, etc. At this time, the mapping information can be managed in a namespace-based hierarchical structure. Here, the namespace-based hierarchical structure may be a hierarchical structure of Factory - Line - Equipment - Module - Logical Tag.

[0081] For example, the above SVID standardization (600) maps logical tags and state variable identifiers (SVID) as shown in [Table 1] below.

[0082] Logical Tag SVID Column LT_001 1000 LOT_ID LT_002 1001 RECIPE_ID LT_003 1002 RPM_VALUE

[0083] In this way, the SVID standardization unit (600) can map one or more state variable identifiers by integrating one or more logical tags with the SVID scheme of the SECS / GEM standard.

[0084] The above semantic group definition unit (700) defines (or sets) semantic groups based on (or based on) the defined logical tag and / or the mapped state variable identifier.

[0085] That is, the semantic group definition unit (700) defines a semantic group by grouping the corresponding logical tags and / or corresponding state variable identifiers according to one or more of a plurality of criteria, such as a set of tags belonging to the same facility, a set of tags used in the same process step, a set of tags related to the same event (CEID: Collection Event ID), a set of tags dependent on the same recipe, a set of tags corresponding to the same LoT ID, etc. Here, the semantic group is a process semantic unit and is a logical set unit that manages multiple logical tags or multiple state variable identifiers (SVIDs) by grouping them together. It is used as a synchronization unit for generating datasets for artificial intelligence learning and includes a group ID, group name, a list of included logical tags, a list of included SVIDs, an applied process step, event linkage information, group version information, etc.

[0086] In this way, the semantic group definition unit (700) can define one or more semantic groups for one or more logical tags and / or one or more state variable identifiers.

[0087] The above dataset schema definition unit (800) predefines (or sets) the structure of the artificial intelligence training dataset to be created. Here, the artificial intelligence training dataset to be created is defined as a table (or column) structure and can be converted into a tensor structure or a vector structure depending on the training model input format, and includes a timestamp, equipment ID, lot ID, process step, SVID group, derived feature value, logical tag value, label value (or label field) (e.g., classification value such as normal, abnormal), version ID, raw value, derived variable, etc.

[0088] That is, the dataset schema definition unit (800) predefines (or sets) the structure of the AI ​​training dataset to be generated, including a learning purpose (e.g., defect prediction, anomaly detection), an AI model type (e.g., Classification Regression), a list of logical tags to be included, definitions of semantic groups to be included, definitions of features (e.g., Raw, Derived, Label), etc.

[0089] At this time, the dataset schema definition unit (800) includes a column definition module (not shown) that defines the column structure of the dataset and a key tag condition definition module (not shown) that defines a key tag condition expression, and has the characteristic of pre-defining the structure and the generation gate in the stage prior to dataset creation. Here, the column definition module includes a column name, data type, whether it is a key, feature type, version ID, etc. Additionally, the key tag condition definition module includes a key tag list, condition expression, evaluation time, record policy, etc.

[0090] Here, the above dataset schema definition (or dataset schema definition for artificial intelligence learning) includes the following two steps.

[0091] 1) Dataset unit definition step

[0092] The corresponding dataset unit definition is a step of defining a logical identifier for the AI ​​training dataset being generated, and includes a dataset name, dataset version, applicable equipment or process identifier, generation conditions (including, for example, events, trigger conditions, etc.), sampling policy information, etc.

[0093] At this time, the above dataset name may be defined based on criteria including the learning purpose, facility unit, process step unit, lot unit, etc.

[0094] For example, the dataset schema definition unit (800) defines dataset units including ETCH_MAINSTEP_ANOMALY_DS, an anomaly detection dataset used in a semiconductor etching process; BATTERY_FORMING_TRACE_DS, a dataset used in a battery manufacturing process data system; EDGE_REALTIME_MONITOR_DS, a dataset used in a smart factory, IoT, and manufacturing data platform.

[0095] 2) Define dataset structure (or define table-based schema)

[0096] The structure of the dataset is defined as a table-based column structure.

[0097] The dataset consists of multiple columns, and each column includes column name, data type, data format, key status, tag mapping information, etc.

[0098] In this case, the column name is defined as a logical field name that identifies the meaning of the corresponding data within the dataset, and includes timestamp, equipment ID, lot ID, process step, temperature, pressure, RF power, label, etc.

[0099] Furthermore, the above data type defines the type of data referenced by the corresponding column and includes System Variables, PLC Tags, SVIDs, Derived Features, User Defined Variables, etc. Here, the System Variables are variables generated at the system level, such as timestamps, Equipment IDs, and Lot IDs. Additionally, the PLC Tag represents a standardized tag defined in the Logical Tag Layer. Furthermore, the SVID represents a SECS / GEM standard state variable identifier. Additionally, the Derived Feature represents a derived variable generated in the AI ​​Feature Definition Layer.

[0100] In addition, the above data format defines the physical storage format of each column and is defined to be directly compatible with the artificial intelligence model input format, and includes Integer, Float, Double, Boolean, String, Enum, Vector, Tensor, etc.

[0101] Additionally, regarding the aforementioned key status, each column can be defined as a Primary Key, Composite Key, or Non-Key field. In this case, the key field includes Timestamp, Equipment ID, Lot ID, Event ID, etc., thereby guaranteeing the uniqueness and temporal alignment structure of the dataset.

[0102] In addition, the above tag mapping information (or column-specific tag mapping) is configured so that each column is mapped to a logical tag or SVID defined in the PLC Map standardization unit (500). At this time, the mapping structure includes 1:1 tag mapping, multi-tag aggregation mapping, conditional mapping, version-based mapping, etc. In addition, each column mapping is performed based on a logical tag ID or SVID rather than a physical address.

[0103] The following [Table 2] shows an example of a dataset schema definition.

[0104] Dataset Name: ETCH_ANOMALY_DS Column Name Data Type Data Format Key Tag Mapping Timestamp System Variable String Y DateTime Equipment_ID System Variable String Y EquipmentName Temperature PLC Tag Float N LOGIC_TEMP_01 Pressure SVID Float N SVID_1023 Temp_MA_5s Derived Feature Float N Derived(LOGIC_TEMP_01) Label User Variable Enum N AL Inference

[0105] In addition, the dataset schema definition unit (800) maps the physical address-based data, logical tags defined in relation to the physical address-based data, state variable identifiers, semantic groups, and the structure of the artificial intelligence learning dataset to be generated, and stores them in the storage unit (200).

[0106] In this way, the dataset schema definition unit (800) can store, for a plurality (or one or more) of physical address-based data, the physical address-based data, the logical tag defined in relation to the physical address-based data, the state variable identifier, the semantic group, the structure of the artificial intelligence learning dataset to be generated, etc., respectively, by mapping them to the storage unit (200).

[0107] The above dataset generating interface driver (or dataset generating interface unit) (900) loads (or collects) one or more data related to the user request from among a plurality of physical address-based data stored in the storage unit (200), according to the structure of the AI ​​learning dataset to be generated previously defined, in accordance with the user request.

[0108] That is, the dataset generating interface driver (900) generates a data snapshot synchronized with one or more data related to the user request in semantic group units among a plurality of physical address-based data stored in the storage unit (200), according to the structure of the AI ​​learning dataset to be generated previously, and converts it into an AI learning dataset structure. Here, the dataset generating interface driver (900) is structured to directly generate a dataset at the interface driver stage, and the interface driver operates based on a logical tag or state variable identifier (SVID) rather than a physical address.

[0109] Thus, the dataset generating interface driver (900) converts one or more data related to the loaded (or collected) corresponding user request into logical tags and state variable identifiers (SVIDs), organizes (or synchronizes) the converted logical tags and state variable identifiers (SVIDs) into semantic groups to generate the data snapshot, and converts the generated data snapshot into the artificial intelligence learning dataset structure.

[0110] At this time, the dataset generation type interface driver (900) generates the artificial intelligence learning dataset by generating a data snapshot synchronized by semantic group unit according to the structure of the artificial intelligence learning dataset to be generated defined in the dataset schema definition unit (800), by inputting a PLC real-time data value, an event occurrence signal (including, for example, a PLC value change, CEID, etc.), semantic group definition information, a snapshot time window setting value, and sampling period information.

[0111] That is, the dataset generating interface driver (900) generates an event-based snapshot (or synchronized data snapshot) including a PLC value change event-based, a CEID (Collection Event ID) generation-based, a user-defined event-based, and a hybrid sampling-based, thereby generating the dataset for artificial intelligence learning. At this time, the dataset generating interface driver (900) performs snapshot synchronization to fix the buffer based on the same PLC Scan Cycle and may include data in the interval before and after the event. Additionally, the user-defined event-based may be information defined by the user regarding what information to read when an event-based trigger is loaded. Furthermore, the hybrid sampling is structured to switch from a first collection cycle in normal circumstances to a second collection cycle when an event occurs, and the second collection cycle is configured to be shorter than the first collection cycle, indicating that sampling is performed for a shorter duration than before when attempting to take a detailed snapshot.

[0112] In this way, the above dataset generating interface driver (900) generates a dataset for artificial intelligence learning according to the following procedure.

[0113] 1. Collect physical address values ​​(or physical address-based data) based on the PLC scan cycle.

[0114] 2. Convert the collected physical address value above into a logical tag layer (or logical tag).

[0115] 3. Map the above converted logical tag to an SVID.

[0116] 4. Organize data buffers by defined semantic group.

[0117] 5. Determination of event occurrence. (Determination of event occurrence based on conditions / criteria related to physical address-based data)

[0118] 6. Fix the time window buffer before and after the event occurrence.

[0119] 7. Generate derived variables based on the AI ​​feature definition layer.

[0120] 8. Serialize into the final dataset format (e.g., including JSON, Tensor, CSV, etc.)

[0121] In this way, the above dataset generating interface driver (900) performs the procedures within the driver without going through a separate central server.

[0122] In addition, the dataset generating interface driver (900) generates a dataset for artificial intelligence learning based on the predefined schema.

[0123] Accordingly, an AI training dataset with a structure already determined at the time of data generation is created.

[0124] The structure has the following effects.

[0125] 1. Prior standardization of dataset structure.

[0126] 2. Ensuring training data consistency.

[0127] 3. Column structure maintained even when changing PLC.

[0128] 4. Dataset structure can be reused even when replacing the AI ​​model.

[0129] 5. Automatic formation of data accumulation structure in a Dark Factory environment.

[0130] In this way, the above dataset generating interface driver (900) generates a data snapshot based on the value of a tag that serves as a trigger for changing a logical tag or SVID value, and generates (or converts) a dataset for artificial intelligence learning.

[0131] For example, when the button named 'start' is clicked on the equipment, the equipment sets the controller values ​​of the equipment using the Set Value of the Recipe for the corresponding task, and actual measurement values ​​are measured on the equipment according to the set values. At this time, when the equipment writes the value '1', which means 'On', to the value of the PLC tag named 'Started', the data set generating interface driver (900) detects the value and reads the tag values ​​of the semantic group linked to the tag to generate the data snapshot.

[0132] The data integrity verification unit (1000) performs pre-set numerical validity verification, state transition-based verification, cross-tag consistency verification, process rule and physical constraint-based verification, etc., on the generated artificial intelligence learning dataset to guarantee the quality of the generated artificial intelligence learning dataset, and calculates (or calculates / estimates) an integrity score (or quality grade). At this time, the data integrity verification unit (1000) is composed of an operation module including arithmetic operations and statistical functions and includes a Value Filter (not shown) that performs a cleaning function (or filtering function), a state transition analyzer (not shown) that performs process state transition graph-based verification, a cross-tag causal analyzer (not shown) that analyzes logical relationships between tags, a physical constraint verifier (not shown) that detects physically impossible combinations, and an integrity score calculator (not shown) that calculates the final integrity score.

[0133] (1) Numerical validity verification

[0134] Here, the numerical validity verification includes minimum value, maximum value, rate of change, moving average, standard deviation, variance, IQR (Interquartile Range), and z-score-based outlier detection using the value filter.

[0135] Additionally, the numerical validity verification includes a type verification that determines whether the data type and data format defined in the dataset schema definition unit (800) match.

[0136] That is, the data integrity verification unit (1000) calculates a numerical validity score (N) (maximum score is 1) by reflecting the minimum value, maximum value, rate of change, moving average, standard deviation, variance, IQR (Interquartile Range), z-score, etc.

[0137] At this time, the rule examples are as follows.

[0138] - Allowable temperature range: 300 ~ 340℃

[0139] - Pressure tolerance: 0.8 ~ 1.5 Torr

[0140] - RF Power Allowable Range: 1400 ~ 1600 W

[0141] - Allowable Gas Flow Range: 45 ~ 55 sccm

[0142] - Allowable difference between Temperature and Temp_MA_5s: ±3.0℃

[0143] - Pressure_Delta Allowed: ≤ 0.10

[0144] - No penalty for no type errors

[0145] In addition, an example of a scoring method (rules defined by actual users) is as follows.

[0146] - Item-specific deductions from initial 1.0:

[0147] Out of range: -0.25

[0148] Excessive rate of change: -0.20

[0149] Moving average deviation exceeded: -0.15

[0150] Type error: -0.40

[0151] - Evaluate current value

[0152] Temperature 321.4 → Normal

[0153] Pressure 1.28 → Normal

[0154] RF Power 1490 → Normal

[0155] Gas Flow 48.0 → Normal

[0156] |321.4 - 320.8| = 0.6 → Normal

[0157] Pressure_Delta 0.03 → Normal

[0158] Type Normal

[0159] Accordingly, the data integrity verification unit (1000) calculates a numerical validity score (N) based on numerical validity verification.

[0160] (2) State transition-based verification

[0161] In addition, the data integrity verification unit (1000) analyzes the pattern of change of tag values ​​according to time order to identify process state transitions.

[0162] In addition, the above state transition-based verification determines only allowed transitions based on a predefined state transition graph as normal states, and determines the corresponding data as abnormal data if an unallowed state transition occurs.

[0163] That is, the data integrity verification unit (1000) determines whether it is normal based on the allowed state transition graph, determines that the unallowed transition is abnormal, and calculates the state transition score (S) (maximum score 1).

[0164] - Allowed transition example

[0165] IDLE → READY

[0166] READY → ETCH_RUN

[0167] ETCH_RUN → ETCH_END

[0168] ETCH_END → READY

[0169] - today:

[0170] Prev_State = READY

[0171] Curr_State = ETCH_RUN

[0172] This is an allowed transition.

[0173] Accordingly, the data integrity verification unit (1000) calculates a state transition score (S) based on state transition-based verification.

[0174] (3) Cross-tag consistency verification

[0175] In addition, the data integrity verification unit (1000) analyzes the correlation between multiple tags or equipment operation constraints to determine logical consistency.

[0176] In addition, the cross-tag consistency verification includes a causal relationship-based verification that determines whether a corresponding physical quantity changes when a control signal is generated.

[0177] That is, the data integrity verification unit (1000) calculates a cross-tag consistency score (C) (maximum score is 1) by including a causal relationship verification, such as whether there is a change in a corresponding physical quantity when a control signal is generated.

[0178] - Rule Example

[0179] If Valve_Open = TRUE, Gas_Flow must be ≥ 45

[0180] If RF_Power > 1450, Process_Step must be MAIN_ETCH

[0181] After CEID = PROCESS_START, Pressure must be established at 0.8 or higher

[0182] - Evaluate current value

[0183] Valve_Open = TRUE, Gas_Flow = 48.0 → Normal

[0184] RF_Power = 1490, Step = MAIN_ETCH → Normal

[0185] CEID = PROCESS_START, Pressure = 1.28 → Normal

[0186] Accordingly, the data integrity verification unit (1000) calculates a cross-tag consistency score (C) based on cross-tag consistency verification.

[0187] (4) Verification based on process rules and physical constraints

[0188] In addition, the data integrity verification unit (1000) detects data combinations that violate predefined process rules or physical constraints.

[0189] The above physical constraints include physical correlations between pressure, temperature, flow rate, or velocity.

[0190] That is, the data integrity verification unit (1000) performs a verification of physical correlations between pressure, temperature, flow rate, and speed to calculate a physical constraint score (P) (maximum score is 1).

[0191] At this time, the rule examples are as follows.

[0192] If Valve_Open = TRUE and Gas_Flow > 45, Pressure must be at least 0.9.

[0193] If the RF_Power is around 1500W, the temperature is at least 310℃ or higher

[0194] A pressure of 0.9 to 1.4 Torr is recommended during the MAIN_ETCH stage.

[0195] - Evaluate current value

[0196] Gas_Flow 48.0 + Valve_Open TRUE + Pressure 1.28 → Normal

[0197] RF_Power 1490 + Temperature 321.4 → Normal

[0198] MAIN_ETCH Pressure 1.28 → Normal

[0199] Accordingly, the data integrity verification unit (1000) calculates a physical constraint score (P) based on process rule and physical constraint-based verification.

[0200] In addition, the data integrity verification unit (1000) calculates the integrity score by performing a numeric validity verification, state transition-based verification, cross-tag consistency verification, process rule and physical constraint-based verification, etc., on the generated artificial intelligence learning dataset as shown in [Equation 1] below, and then multiplying the numeric validity score (N), state transition score (S), cross-tag consistency score (C), and physical constraint score (P), each calculated by the respective values, by a pre-set weight and summing them.

[0201]

[0202] Here, the above weights use a weighted sum and may be w1=0.35, w2=0.20, w3=0.25, w4=0.20.

[0203] In addition, the aforementioned integrity score is included as metadata in the dataset and is used as a weight or exclusion criterion when training the artificial intelligence model.

[0204] In addition, the rule identifier and rule version information used in the above verification process are stored together to ensure the reproducibility of the data generation history.

[0205] The key tag condition evaluation unit (1100) checks (or determines) whether to create a record (or whether to create a record according to the record creation condition) for the generated artificial intelligence learning dataset based on the record creation condition for the key tag defined in the dataset schema definition unit (800). Here, the key tag includes one or more tags that constitute the primary key or composite key of the dataset. Additionally, the record creation condition is defined as a Boolean condition expression, a comparison operation condition, a range condition, an event occurrence condition, or a combination of these conditions.

[0206] That is, the key tag condition evaluation unit (1100) comprises a key tag set (not shown) composed of tags corresponding to the primary key or composite key of the dataset, a condition expression unit (not shown) that interprets a condition expression, and a condition evaluator (not shown) that determines the true or false nature of the condition. In this case, if the condition is true, record creation is allowed.

[0207] At this time, examples of the above record generation conditions are as follows.

[0208] Tag_A == TRUE

[0209] Tag_B > Threshold

[0210] Tag_C Change Rate >= Setting Value

[0211] Whether Event_ID occurred

[0212] - AND / OR combination conditions of multiple tags

[0213] In addition, the key tag condition evaluation unit (1100) configures a condition-based generation structure to prevent excessive accumulation of unnecessary normal data and to accumulate only meaningful event-centered data as training data.

[0214] In addition, the key tag condition evaluation unit (1100) is configured to simultaneously evaluate multiple tags included in the composite key for a composite key-based condition.

[0215] For example, you can configure it to create a record only when the Lot_ID is valid, the process stage is a specific stage, and the alarm tag is TRUE.

[0216] If, based on the above verification result (or above judgment result), it is confirmed that record generation is impossible for the generated artificial intelligence learning dataset based on the record generation conditions for the key tag defined in the dataset schema definition unit (800), the key tag condition evaluation unit (1100) may not generate the data (or artificial intelligence learning dataset) at that time as a dataset record, and may selectively store or discard the data at that time in a buffer.

[0217] That is, if, as a result of the above verification (or the above judgment), at least one of the record creation conditions for the key tag defined in the dataset schema definition unit (800) is not satisfied (or is not satisfied), the key tag condition evaluation unit (1100) may not create the data at that time as a dataset record, and may selectively store or discard the data at that time in a buffer.

[0218] In addition, if, based on the above verification result (or above judgment result), it is confirmed that record generation is possible for the generated artificial intelligence learning dataset based on the record generation conditions for the key tag defined in the dataset schema definition unit (800), the process can proceed to the next process (for example, the processing process in the threshold comparison unit).

[0219] The threshold comparison unit (1200) determines (or verifies) whether the integrity score calculated (or calculated / calculated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and whether the generated artificial intelligence learning dataset satisfies the record generation condition in the key tag condition evaluation unit (1100), respectively.

[0220] In this way, the threshold comparison unit (1200) can determine whether to record the corresponding artificial intelligence learning dataset.

[0221] When the integrity score calculated (or calculated / calculated) by the data integrity verification unit (1000) is less than a preset threshold value based on the above judgment result (or above verification result), and when at least one of the cases applies where the generated artificial intelligence learning dataset does not satisfy the record generation condition in the key tag condition evaluation unit (1100) (or records are not generated), the threshold comparison unit (1200) generates verification failure data and separates and stores the generated verification failure data in a preset isolation storage within the storage unit (200).

[0222] Additionally, if the integrity score calculated (or calculated / estimated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and if the generated artificial intelligence learning dataset satisfies the record generation condition in the key tag condition evaluation unit (1100) (or record generation), then the threshold comparison unit (1200) requests record generation from the record generation unit (1300).

[0223] That is, if, based on the above judgment result (or the above verification result), the integrity score calculated (or calculated / calculated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and the generated artificial intelligence learning dataset satisfies the record creation condition in the key tag condition evaluation unit (1100) (or record creation), the threshold comparison unit (1200) transmits (or sends) record creation request information, etc. to the record creation unit (1300).

[0224] In addition, the above threshold comparison unit (1200) may also be processed according to the following operating criteria.

[0225] - Operating Standards

[0226] Based on the score, whether to create a record or through labeling can serve as a condition for future queries by Dataset Grade.

[0227] 90 ~ 100: A, 100% reflection of learning

[0228] 75 ~ 89: B, Learning Reflection

[0229] 60 ~ 74: C, Low weighting applied

[0230] Less than 60: D, excluded from learning

[0231] Create final Dataset Record only when 80 or higher

[0232] In an embodiment of the present invention, the threshold comparison unit (1200) primarily describes determining simultaneously whether the integrity score calculated (or calculated / estimated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and whether the generated artificial intelligence learning dataset satisfies the record generation condition in the key tag condition evaluation unit (1100); however, the invention is not limited thereto, and the threshold comparison unit (1200) may be configured to determine only one of the conditions, whether the integrity score calculated (or calculated / estimated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and whether the generated artificial intelligence learning dataset satisfies the record generation condition in the key tag condition evaluation unit (1100).

[0233] That is, the threshold comparison unit (1200) can determine whether the integrity score calculated (or calculated / calculated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, or whether the generated artificial intelligence learning dataset satisfies the record generation condition in the key tag condition evaluation unit (1100).

[0234] The record generation unit (1300) stores the previously generated data snapshot, the previously generated artificial intelligence training dataset, etc., in the storage unit (200) in response to a request from the threshold comparison unit (1200).

[0235] That is, the record generation unit (1300) receives record generation request information transmitted from the threshold comparison unit (1200), and based on the received record generation request information, stores a previously generated data snapshot, a previously generated artificial intelligence training dataset, etc. in the storage unit (200).

[0236] Additionally, when one of the AI ​​learning datasets is selected from a list corresponding to AI learning datasets stored in the storage unit (200) according to user selection (or user input / touch / control) and a dataset format is selected (or set), the serializer (1400) converts the selected AI learning dataset into the selected (or set) dataset format. Here, the dataset format includes JSON, Tensor, CSV, vector array, etc.

[0237] The above integrated control unit (controller, or MCU (microcontroller unit)) (1500) executes the overall control function of the device (10) that generates the artificial intelligence learning dataset at the interface driver stage.

[0238] Additionally, the integrated control unit (1500) executes the overall control function of the device (10) that generates an artificial intelligence learning dataset at the interface driver stage using the program and data stored in the storage unit (200). The integrated control unit (1500) may include RAM, ROM, CPU, GPU, and a bus, and the RAM, ROM, CPU, GPU, etc. may be connected to each other through a bus. The CPU can access the storage unit (200) and perform booting using the O / S stored in the storage unit (200), and can perform various operations using various programs, content, data, etc. stored in the storage unit (200).

[0239] The integrated control unit (1500) transmits the AI ​​learning dataset converted into the dataset format to an AI server (not shown) through the communication unit (100).

[0240] Accordingly, the AI ​​server performs LLM (Large Language Model)-based learning (or artificial neural network / machine learning / deep learning) based on the AI ​​training dataset converted into the corresponding dataset format, and generates (or predicts / classifies) feedback as a learning result.

[0241] That is, the AI ​​server performs learning (or inference / AI / machine learning / deep learning) using an AI training dataset converted into the corresponding dataset format as an input value (or prompt) for a pre-configured large-scale language model, and generates (or predicts / classifies) feedback as a learning result (or inference result / AI result / machine learning result / deep learning result / completion). Here, the feedback is fed back to the equipment via methods such as PLC command write-back, SECS SVID update, CEID trigger, or MES interface, thereby forming an autonomous process control structure.

[0242] In this way, by feeding back the AI ​​judgment result (or feedback) to the PLC or SVID layer in real time, a closed-loop control structure can be formed (or a system that measures output / results through sensors, compares them with inputs / targets, and automatically modifies inputs through a controller to reduce errors).

[0243] Accordingly, the facility can provide autonomous process optimization without an external analysis server.

[0244] In addition, the artificial intelligence server transmits the generated feedback (or learning result / inference result) to the communication unit (100).

[0245] Additionally, the communication unit (100) receives feedback transmitted from the artificial intelligence server (or feedback corresponding to an artificial intelligence learning dataset converted into the corresponding dataset format) under the control of the integrated control unit (1500).

[0246] In addition, the integrated control unit (1500) works in conjunction with the SVID standardization unit (600) to perform an SVID update function (or an SVID update function per logical tag), etc., based on the received feedback.

[0247] In addition, the integrated control unit (1500) transmits the received feedback to the equipment (or equipment control layer) through the communication unit (100) in order to control the equipment (or PLC) based on the received feedback.

[0248] Accordingly, the above equipment (or equipment control layer) performs PLC write-back functions, event trigger functions, legacy system interface (MES) functions, etc., based on the above feedback.

[0249] In this way, the AI ​​training dataset converted into the above dataset format can be used for the AI ​​model training stage, the AI ​​model validation stage, real-time inference input data, edge AI model update, closed-loop control input, etc.

[0250] Additionally, when a project file is uploaded to a device (10) that generates an artificial intelligence learning dataset at the interface driver stage according to user input (or user selection / touch / control) (or when PLC program change information is uploaded), the integrated control unit (1500) is configured (or managed) to maintain the logical tag and schema structure even if the physical address related to the PLC is changed by linking with the PLC MAP standardization unit (500) and the dataset schema definition unit (800).

[0251] In addition, when the physical address changes, the integrated control unit (1500) performs an automatic remapping function in conjunction with the PLC MAP standardization unit (500) and the dataset schema definition unit (800).

[0252] That is, when a change to the corresponding PLC program is detected, the integrated control unit (1500) parses the corresponding PLC program, checks for a change in the physical address based on the parsed PLC program, and then determines whether to maintain the logical tag.

[0253] If, based on the above judgment result, it is determined that the logical tag is maintained, the integrated control unit (1500) terminates the entire process.

[0254] In addition, if it is determined that the logical tag is to be changed based on the above judgment result, the integrated control unit (1500) is configured (or managed) to maintain the corresponding logical tag and schema structure through administrator approval, or is configured (or managed) to perform automatic remapping and update the map version.

[0255] In this way, the device (10) that generates an artificial intelligence learning dataset at the interface driver stage establishes (or provides / configures) a system for pre-defining the artificial intelligence learning structure at a stage prior to data generation, unlike conventional technology that assigns meaning after data collection, by pre-defining the structure of the dataset through the dataset schema definition unit (800).

[0256] Accordingly, it provides effects such as ensuring consistency in data interpretation, maintaining structural stability when the process is changed, and enabling structural reuse when the facility is expanded.

[0257] In addition, the device (10) that generates the artificial intelligence learning dataset at the interface driver stage automatically (or directly) generates the dataset according to a predefined schema at the interface driver stage without going through a central data lake or ETL (Extract Transform Load) reprocessing stage.

[0258] Accordingly, it provides benefits such as simplified system architecture, reduced server load, minimized data latency, and the ability to apply real-time artificial intelligence.

[0259] In addition, the device (10) that generates an artificial intelligence learning dataset at the interface driver stage provides effects such as ensuring semantic continuity with past data, maintaining a long-term learning model, and reducing retraining costs, because the logical tag and schema structure are maintained even if the physical address related to the PLC is changed when the PLC program is changed.

[0260] In addition, the device (10) that generates a dataset for artificial intelligence learning at the interface driver stage provides effects such as precise collection of the interval before and after an event, reduction of noise data, optimization of training data density, and improvement of anomaly detection accuracy through an event-based snapshot and hybrid sampling structure.

[0261] In addition, the device (10) that generates an artificial intelligence learning dataset at the interface driver stage forms a PLC-SECS-MES-AI integrated data model, integrates PLC logic tags and SVIDs, and links them with a dataset schema, thereby providing effects such as forming a single data model of the equipment control layer and the AI ​​analysis layer, implementing a closed-loop control structure, and securing a basis for autonomous process optimization.

[0262] In addition, the device (10) that generates the artificial intelligence learning dataset in the interface driver stage performs range error removal, change rate anomaly detection, statistical outlier removal, type mismatch verification, etc. through a Value Filter including arithmetic operations and statistical functions.

[0263] Accordingly, this provides the effect of generating only numerically reliable training data, rather than simple data accumulation.

[0264] In addition, the device (10) that generates the artificial intelligence learning dataset at the interface driver stage performs structural verification such as state transition graph-based verification, tag-to-tag causal relationship-based verification, and physical constraint-based verification.

[0265] Accordingly, it provides effects such as removing data that violates process logic, excluding physically impossible data, and blocking process step error data.

[0266] This is a process meaning-based integrity assurance effect that is qualitatively distinct from a simple preprocessing role.

[0267] In addition, the device (10) that generates a dataset for artificial intelligence learning in the interface driver stage calculates a Semantic Integrity Score for each data sample and includes the calculated integrity score in the dataset, thereby providing effects such as the ability to reflect weights during AI learning, the ability to automatically exclude low-quality data, improved model prediction stability, and ensure reproducibility.

[0268] In addition, the device (10) that generates an artificial intelligence learning dataset in the interface driver stage as described above sets a condition on a key tag constituting the primary key or composite key of the dataset and generates a record only when the condition is true, thereby providing effects such as preventing excessive accumulation of unnecessary normal data, accumulating event-centered data, improving storage efficiency, and forming a learning structure centered on meaningful process change intervals.

[0269] In addition, since conditions can be set in combination with integrity scores, it provides the effect of generating a dataset only from high-quality data that has passed both numerical and semantic verification.

[0270] In addition, the device (10) that generates a dataset for artificial intelligence learning at the interface driver stage provides effects such as continuous accumulation of learning data, real-time decision-making by edge artificial intelligence, and formation of a foundation for autonomous process control by designing a 'structure in which data is bound to accumulate' in advance.

[0271] This has significant industrial value in that it provides a data infrastructure structure (or structural foundation) for the implementation of a Dark Factory.

[0272] In addition, the device (10) that generates an artificial intelligence learning dataset at the interface driver stage provides a hierarchical protection structure capable of preventing bypass design.

[0273] In other words, the multi-layered PLC Map structure, SVID integration structure, schema definition section, direct dataset generation structure within the driver, value filter-based verification structure, and key tag condition-based record generation structure are organically combined, so that the full effect cannot be achieved if only some elements are separated and implemented, thereby forming a defensive structure that is difficult to bypass.

[0274] In addition, the device (10) that generates a dataset for artificial intelligence learning in the interface driver step can be applied to semiconductor equipment.

[0275] - Generate CEID-based process completion event time dataset

[0276] - Temperature, pressure, RPM SVID integration

[0277] - Training a defect prediction model

[0278] In addition, the device (10) that generates a dataset for artificial intelligence learning in the interface driver step can be applied to a secondary battery process.

[0279] - Schema group organization by process stage

[0280] - Hybrid sampling applied

[0281] - PLC Write-back control after anomaly detection

[0282] In addition, the device (10) that generates a dataset for artificial intelligence learning in the interface driver stage can be applied to the Edge AI structure.

[0283] - Lightweight model built into the driver

[0284] - Immediate judgment and SVID feedback

[0285] In this way, physical address-based data collected in heterogeneous PLC environments can be standardized into a multi-layered structure, and the standardized physical address-based data can be integrated with the state variable identifier system of the SECS / GEM standard to directly generate a dataset for artificial intelligence training at the interface driver stage.

[0286] In addition, in this way, physical address-based data can be pre-normalized into a multi-layered standardization structure.

[0287] In addition, logical tags can be integrated with the SVID scheme of the SECS / GEM standard in this way.

[0288] In addition, in this way, a dataset for artificial intelligence training can be generated directly at the interface driver stage.

[0289] In addition, in this way, a structure can be provided that maintains data continuity even when the PLC program is changed.

[0290] In addition, high-quality training data can be secured through event-based synchronization and a hybrid sampling structure.

[0291] Hereinafter, a method for generating an artificial intelligence learning dataset in the interface driver step according to the present invention will be described in detail with reference to FIGS. 1 to 3.

[0292] FIGS. 2 and 3 are flowcharts illustrating a method for generating an artificial intelligence learning dataset in the interface driver step according to an embodiment of the present invention.

[0293] First, the communication unit (100) collects (or receives) one or more physical address-based data from a PLC (not shown) within the equipment (not shown). Here, the physical address-based data is the entire data managed by the PLC, and is a value stored in the PLC memory area (e.g., Bit, Word, Double Word, etc.), and includes variables used for equipment control and status representation, varies depending on the type of equipment and the PLC program design, and includes process data such as temperature values, pressure values, Real Time values ​​of equipment utility controllers (e.g., Flow Rate, RPM, Voltage, Current, Cylinder Position Value, etc.), valve open / closed status (e.g., On / Off Bit), process step codes, equipment operation status codes (e.g., Run / Idle / Alarm), alarm flag bits, timer values, counter values, Recipe ID, Lot ID (e.g., when managed by the PLC).

[0294] For example, the first communication unit (100) collects one or more physical address-based data from the first PLC in the semiconductor equipment (S210).

[0295] Subsequently, the PLC MAP standardization unit (or PLC MAP standardization engine) (500) converts the collected physical address-based data into a PLC Map of a preset multi-layer structure and defines (or sets) a Logical Tag. Here, the preset multi-layer structure includes a physical address layer, a Logical Tag layer, an SVID standardization layer, a semantic group layer, an artificial intelligence feature definition layer, etc. At this time, the Physical Address Layer preserves the collected physical address-based data as is and includes a configuration in units of bit, word, and double word. In addition, the Logical Tag Layer converts the physical address into a process semantic-based tag name and expands it into multiple logical tags according to process conditions, even for the same physical address. In addition, the SVID Standardization Layer maps the logical tag to an SVID in order to integrate the PLC and SECS data into the same identification system. In this case, the mapping (or SVID mapping) includes 1:1 mapping, 1:N mapping, N:1 aggregation mapping that aggregates multiple logical tags into a single SVID, conditional mapping, etc. Additionally, the Semantic Group Layer organizes semantic groups into units such as equipment status, process step, event, recipe, and Lot (or Work order). Furthermore, the AI ​​Feature Definition Layer includes AI feature definitions that include physical address values ​​(or raw values / physical address-based data), derived variables, time interval statistics, event-based labels, normal / abnormal classification tags, etc. Here, the derived variables include moving averages, change amounts, standard deviations, derivatives, etc.

[0296] For example, the first PLC MAP standardization unit (500) converts the collected physical address-based data into the multi-layered PLC Map and defines the first logical tag (S220).

[0297] Subsequently, the SVID standardization unit (or SVID standardization layer) (600) maps the logical tag to a status variable identifier (SVID: Status Variable ID) of the SECS / GEM (Semiconductor Equipment Communication Standard / Generic Equipment Model) standard.

[0298] That is, the SVID standardization unit (600) identifies a state variable identifier corresponding to a logical tag among the dictionary (or SVID by logical tag) for SVIDs stored in advance in the storage unit (200), and maps the identified state variable identifier corresponding to the logical tag to the logical tag. Here, the mapping includes conditional rule-based mapping, mapping that changes dynamically according to a recipe or process state, version control table, etc. At this time, the mapping information can be managed in a namespace-based hierarchical structure. Here, the namespace-based hierarchical structure may be a hierarchical structure of Factory - Line - Equipment - Module - Logical Tag.

[0299] For example, the first SVID standardization unit (600) identifies a first state variable identifier (SVID) corresponding to the first logical tag defined above among the SVIDs for each logical tag stored in advance in the first storage unit (200), and maps the first logical tag to the identified first state variable identifier (S230).

[0300] Subsequently, the semantic group definition unit (700) defines (or sets) semantic groups based on (or based on) the defined logical tags and / or the mapped state variable identifiers.

[0301] That is, the semantic group definition unit (700) defines a semantic group by grouping the corresponding logical tags and / or corresponding state variable identifiers according to one or more of a plurality of criteria, such as a set of tags belonging to the same facility, a set of tags used in the same process step, a set of tags related to the same event (CEID: Collection Event ID), a set of tags dependent on the same recipe, a set of tags corresponding to the same LoT ID, etc. Here, the semantic group is a process semantic unit and is a logical set unit that manages multiple logical tags or multiple state variable identifiers (SVIDs) by grouping them together. It is used as a synchronization unit for generating datasets for artificial intelligence learning and includes a group ID, group name, a list of included logical tags, a list of included SVIDs, an applied process step, event linkage information, group version information, etc.

[0302] For example, the first semantic group definition unit (700) defines a first semantic group by grouping the first logical tag or the first state variable identifier into one or more of the plurality of criteria (S240).

[0303] Subsequently, the Dataset Schema Definition Unit (800) predefines (or sets) the structure of the AI ​​training dataset to be created. Here, the AI ​​training dataset to be created is defined as a table (or column) structure and can be converted into a tensor structure or a vector structure depending on the training model input format, and includes a timestamp, equipment ID, lot ID, process step, SVID group, derived feature value, logical tag value, label value (or label field) (e.g., classification value such as normal, abnormal), version ID, raw value, derived variable, etc.

[0304] That is, the dataset schema definition unit (800) predefines (or sets) the structure of the AI ​​training dataset to be generated, including a learning purpose (e.g., defect prediction, anomaly detection), an AI model type (e.g., Classification Regression), a list of logical tags to be included, definitions of semantic groups to be included, definitions of features (e.g., Raw, Derived, Label), etc.

[0305] In addition, the dataset schema definition unit (800) maps the physical address-based data, logical tags defined in relation to the physical address-based data, state variable identifiers, semantic groups, and the structure of the artificial intelligence learning dataset to be generated, and stores them in the storage unit (200).

[0306] For example, the first dataset schema definition unit (800) defines the structure of the artificial intelligence learning dataset to be generated.

[0307] In addition, the first dataset schema definition unit maps the collected one or more physical address-based data, a first logical tag defined in relation to the physical address-based data, a first state variable identifier, a first semantic group, and the structure of the artificial intelligence learning dataset to be generated, and stores them in the first storage unit (S250).

[0308] Subsequently, the dataset generating interface driver (or dataset generating interface unit) (900) loads (or collects) one or more data related to the user request from among a plurality of physical address-based data stored in the storage unit (200), according to the structure of the AI ​​learning dataset to be generated previously defined, in accordance with the user request.

[0309] That is, the dataset generating interface driver (900) generates a data snapshot synchronized with one or more data related to the user request in semantic group units among a plurality of physical address-based data stored in the storage unit (200), according to the structure of the AI ​​learning dataset to be generated previously, and converts it into an AI learning dataset structure. Here, the dataset generating interface driver (900) is structured to directly generate a dataset at the interface driver stage, and the interface driver operates based on a logical tag or state variable identifier (SVID) rather than a physical address.

[0310] For example, the first dataset generating interface driver (900), in response to a first user request, synchronizes the first data related to the first user request in units of semantic groups according to the structure of the artificial intelligence learning dataset to be generated among a plurality of physical address-based data previously stored in the first storage unit, and generates the following first data snapshot, thereby converting it into a first artificial intelligence learning dataset.

[0311] {

[0312] "Timestamp": "2026-03-10T10:32:16.021",

[0313] "Equipment_ID": "ETCH_01",

[0314] "Lot_ID": "L20260310001",

[0315] "Process_Step": "MAIN_ETCH",

[0316] "Valve_Open": true,

[0317] "Gas_Flow": 12.0,

[0318] "Pressure": 0.42,

[0319] "Temperature": 352.0,

[0320] "RF_Power": 1490,

[0321] "Temp_MA_5s": 321.0,

[0322] "Pressure_Delta": 0.22,

[0323] "Prev_State": "READY",

[0324] "Curr_State": "ETCH_END",

[0325] "CEID": "PROCESS_START"

[0326] }(S260).

[0327] Subsequently, the data integrity verification unit (1000) performs a pre-set numerical validity verification, state transition-based verification, cross-tag consistency verification, process rule and physical constraint-based verification, etc., on the generated artificial intelligence training dataset to guarantee the quality of the generated artificial intelligence training dataset, and calculates (or calculates / estimates) an integrity score (or quality grade).

[0328] That is, the data integrity verification unit (1000) calculates the integrity score by performing a numeric validity verification, a state transition-based verification, a cross-tag consistency verification, a process rule and physical constraint-based verification, etc., on the generated artificial intelligence learning dataset as described in [Equation 1] above, and then multiplying the numeric validity score (N), the state transition score (S), the cross-tag consistency score (C), and the physical constraint score (P), each calculated by the respective values, by a pre-set weight and summing them.

[0329] For example, the first data integrity verification unit (1000) performs the following numerical validity verification on the generated first artificial intelligence learning dataset,

[0330] Temperature 352.0 → Out of range (-0.25)

[0331] Pressure 0.42 → Exceeds range (-0.25)

[0332] Pressure_Delta 0.22 → Exceeds rate of change (-0.20)

[0333] |352.0 - 321.0| = 31.0 → Exceeds Moving Average Deviation (-0.15)

[0334] Accordingly, the numerical validity score (N) = 1.00 - 0.25 - 0.25 - 0.20 - 0.15 = 0.15 is calculated.

[0335] In addition, the first data integrity verification unit performs the following state transition-based verification on the generated first artificial intelligence training dataset,

[0336] Assuming READY → ETCH_END is not allowed

[0337] Accordingly, the state transition score (S) = 0 is calculated.

[0338] In addition, the first data integrity verification unit performs the following cross-tag consistency verification on the generated first artificial intelligence training dataset,

[0339] Valve_Open = TRUE but Gas_Flow = 12.0 → Failure

[0340] Pressure less than 0.8 immediately after PROCESS_START → Failure

[0341] RF_Power is 1490 but process formation pressure is too low → Failure

[0342] Accordingly, if we consider 0 out of 3 as successes,

[0343] Calculate the cross-tag consistency score (C) = 0.00.

[0344] At this time, if zero is too strong, a bottom value such as 0.1 can be set.

[0345] In addition, the first data integrity verification unit performs the following process rule and physical constraint-based verification on the generated first artificial intelligence training dataset,

[0346] There is almost no gas flow, but the RF power is high and the pressure is too low.

[0347] Temperature is excessively high, but process pressure is not formed

[0348] Accordingly, the physical constraint score (P) = 0.10 is calculated.

[0349] In addition, the first data integrity verification unit applies the calculated numerical validity score (N=0.15), the calculated state transition score (S=0.00), the calculated cross-tag consistency score (C=0.00), and the calculated physical constraint score (P=0.10), along with preset weights w1=0.35, w2=0.20, w3=0.25, and w4=0.20, to the preceding [Equation 1],

[0350] Integrity score = 100 * (0.35*0.15 + 0.20*0.00 + 0.25*0.00 + 0.20*0.10)

[0351] Calculate = 100 * (0.0525 + 0 + 0 + 0.02) = 7.25 (S270).

[0352] Subsequently, the key tag condition evaluation unit (1100) checks (or determines) whether to create a record (or whether to create a record according to the record creation condition) for the generated artificial intelligence learning dataset based on the record creation condition for the key tag defined in the dataset schema definition unit (800). Here, the key tag includes one or more tags that constitute the primary key or composite key of the dataset. In addition, the record creation condition is defined as a Boolean condition expression, a comparison operation condition, a range condition, an event occurrence condition, or a combination of these conditions.

[0353] That is, the key tag condition evaluation unit (1100) comprises a key tag set (not shown) composed of tags corresponding to the primary key or composite key of the dataset, a condition expression unit (not shown) that interprets a condition expression, and a condition evaluator (not shown) that determines the true or false nature of the condition. In this case, if the condition is true, record creation is allowed.

[0354] For example, the first key tag condition evaluation unit (1100) checks whether to create a record for the first artificial intelligence learning dataset created based on the record creation condition for the key tag defined in the first dataset schema definition unit (S280).

[0355] Subsequently, the threshold comparison unit (1200) determines (or verifies) whether the integrity score calculated (or calculated / calculated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and whether the generated artificial intelligence learning dataset satisfies the record generation condition in the key tag condition evaluation unit (1100), respectively.

[0356] For example, the first threshold comparison unit (1200) determines whether the first integrity score calculated by the first data integrity verification unit is greater than or equal to a preset first threshold, and whether the first artificial intelligence learning dataset satisfies the record generation condition in the first key tag condition evaluation unit (S290).

[0357] When the integrity score calculated (or calculated / calculated) by the data integrity verification unit (1000) is less than a preset threshold value based on the above judgment result (or above verification result), and when at least one of the cases applies where the generated artificial intelligence learning dataset does not satisfy the record generation condition in the key tag condition evaluation unit (1100) (or records are not generated), the threshold comparison unit (1200) generates verification failure data and separates and stores the generated verification failure data in a preset isolation storage within the storage unit (200).

[0358] For example, when the record generation condition in the first key tag condition evaluation unit is not satisfied by the first artificial intelligence learning dataset as a result of the above judgment, the first threshold comparison unit generates first verification failure data and stores the generated first verification failure data in a preset first isolation storage unit within the first storage unit (S300).

[0359] Additionally, if the integrity score calculated (or calculated / estimated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and if the generated artificial intelligence learning dataset satisfies the record generation condition in the key tag condition evaluation unit (1100) (or record generation), then the threshold comparison unit (1200) requests record generation from the record generation unit (1300).

[0360] That is, if, based on the above judgment result (or the above verification result), the integrity score calculated (or calculated / calculated) by the data integrity verification unit (1000) is greater than or equal to a preset threshold, and the generated artificial intelligence learning dataset satisfies the record creation condition in the key tag condition evaluation unit (1100) (or record creation), the threshold comparison unit (1200) transmits (or sends) record creation request information, etc. to the record creation unit (1300).

[0361] For example, when, as a result of the above judgment, the first integrity score calculated by the first data integrity verification unit is greater than or equal to a preset first threshold, and the first artificial intelligence learning dataset satisfies the record creation condition in the first key tag condition evaluation unit, the first threshold comparison unit transmits the first record creation request information to the first record creation unit (1300) (S310).

[0362] Subsequently, the record generation unit (1300) stores the previously generated data snapshot, the previously generated artificial intelligence training dataset, etc., in the storage unit (200) in response to a request from the threshold comparison unit (1200).

[0363] That is, the record generation unit (1300) receives record generation request information transmitted from the threshold comparison unit (1200), and based on the received record generation request information, stores a previously generated data snapshot, a previously generated artificial intelligence training dataset, etc. in the storage unit (200).

[0364] For example, the first record generation unit receives first record generation request information transmitted from the first threshold comparison unit, and stores the generated first artificial intelligence learning dataset, etc., in the first storage unit based on the received first record generation request information (S320).

[0365] Subsequently, when one of the AI ​​learning datasets is selected from a list corresponding to AI learning datasets stored in the storage unit (200) according to user selection (or user input / touch / control) and a dataset format is selected (or set), the serializer (1400) converts the selected AI learning dataset into the selected (or set) dataset format. Here, the dataset format includes JSON, Tensor, CSV, vector array, etc.

[0366] For example, when a first artificial intelligence learning dataset is selected from a list containing information about a plurality of artificial intelligence learning datasets stored in the first storage unit according to a first user selection and a JSON format is selected, the first serialization unit (1400) converts the selected first artificial intelligence learning dataset into the selected JSON format (S330).

[0367] Subsequently, the integrated control unit (1500) transmits the AI ​​learning dataset converted into the above dataset format to an AI server (not shown) through the communication unit (100).

[0368] Accordingly, the AI ​​server performs LLM (Large Language Model)-based learning (or artificial neural network / machine learning / deep learning) based on the AI ​​training dataset converted into the corresponding dataset format, and generates (or predicts / classifies) feedback as a learning result.

[0369] That is, the AI ​​server performs learning (or inference / AI / machine learning / deep learning) using an AI training dataset converted into the corresponding dataset format as an input value (or prompt) for a pre-configured large-scale language model, and generates (or predicts / classifies) feedback as a learning result (or inference result / AI result / machine learning result / deep learning result / completion). Here, the feedback is fed back to the equipment via methods such as PLC command write-back, SECS SVID update, CEID trigger, or MES interface, thereby forming an autonomous process control structure.

[0370] In addition, the artificial intelligence server transmits the generated feedback (or learning result / inference result) to the communication unit (100).

[0371] In addition, the communication unit (100) receives feedback transmitted from the artificial intelligence server under the control of the integrated control unit (1500).

[0372] In addition, the integrated control unit (1500) works in conjunction with the SVID standardization unit (600) to perform an SVID update function (or an SVID update function per logical tag), etc., based on the received feedback.

[0373] For example, the first integrated control unit (1500) transmits the first artificial intelligence learning dataset converted into the JSON format to the artificial intelligence server through the first communication unit, and receives the first feedback transmitted from the artificial intelligence server in response to the transmission.

[0374] In addition, the first SVID standardization unit links with the first integrated control unit to update the SVID for each logical tag based on the received first feedback (S340).

[0375] Subsequently, the integrated control unit (1500) transmits the received feedback to the equipment (or equipment control layer) through the communication unit (100) in order to control the equipment (or PLC) based on the received feedback.

[0376] Accordingly, the above equipment (or equipment control layer) performs PLC write-back functions, event trigger functions, legacy system interface (MES) functions, etc., based on the above feedback.

[0377] In this way, the AI ​​training dataset converted into the above dataset format can be used for the AI ​​model training stage, the AI ​​model validation stage, real-time inference input data, edge AI model update, closed-loop control input, etc.

[0378] For example, the first integrated control unit transmits the received first feedback to the first facility through the first communication unit in order to control the first facility based on the received first feedback.

[0379] In addition, the first facility receives the first feedback transmitted from the first integrated control unit and performs a PLC Write-back function based on the received first feedback (S350).

[0380] Additionally, when a project file is uploaded to a device (10) that generates an artificial intelligence learning dataset at the interface driver stage according to user input (or user selection / touch / control) (or when PLC program change information is uploaded), the integrated control unit (1500) is configured (or managed) to maintain the logical tag and schema structure even if the physical address related to the PLC is changed by linking with the PLC MAP standardization unit (500) and the dataset schema definition unit (800).

[0381] Accordingly, semantic continuity between past and present data can be maintained.

[0382] In addition, when the physical address changes, the integrated control unit (1500) performs an automatic remapping function in conjunction with the PLC MAP standardization unit (500) and the dataset schema definition unit (800).

[0383] That is, when a change to the corresponding PLC program is detected, the integrated control unit (1500) parses the corresponding PLC program, checks for a change in the physical address based on the parsed PLC program, and then determines whether to maintain the logical tag.

[0384] If, based on the above judgment result, it is determined that the logical tag is maintained, the integrated control unit (1500) terminates the entire process.

[0385] In addition, if it is determined that the logical tag is to be changed based on the above judgment result, the integrated control unit (1500) is configured (or managed) to maintain the corresponding logical tag and schema structure through administrator approval, or is configured (or managed) to perform automatic remapping and update the map version.

[0386] In this way, the integrated control unit (1500) can provide a structure that maintains data continuity even when changing the PLC program.

[0387] For example, when a project file is uploaded according to user input, the first integrated control unit is configured to maintain the previously defined logical tag and schema structure by linking with the first PLC MAP standardization unit and the dataset schema definition unit (S360).

[0388] As described above, an embodiment of the present invention standardizes physical address-based data collected in a heterogeneous PLC environment into a multi-layered structure and integrates the physical address-based data standardized into a multi-layered structure with the state variable identifier scheme of the SECS / GEM standard to directly generate a dataset for artificial intelligence learning at the interface driver stage, thereby enabling the generation of a semantic-based AI learning structure at the data generation point (Driver layer) without undergoing a separate data lake and / or ETL (Extract-Transform-Load) reprocessing stage after the collection of physical address-based data.

[0389] In addition, as described above, embodiments of the present invention can form a semantic-based logical tag system by pre-normalizing physical address-based data into a multi-layer standardization structure.

[0390] In addition, as previously described, embodiments of the present invention can construct a PLC-SECS-AI integrated data model by integrating logic tags with the SVID scheme of the SECS / GEM standard.

[0391] In addition, as described above, an embodiment of the present invention can eliminate the post-processing step by directly generating a dataset for artificial intelligence learning at the interface driver stage.

[0392] In addition, as described above, the embodiment of the present invention provides a structure that maintains data continuity even when the PLC program is changed, thereby ensuring data continuity.

[0393] In addition, as described above, embodiments of the present invention can optimize training data density by securing high-quality training data through event-based synchronization and a hybrid sampling structure.

[0394] A person skilled in the art to which the present invention pertains will be able to make modifications and variations to the foregoing without departing from the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are intended to explain, not limit, the technical concept of the present invention, and the scope of the technical concept of the present invention is not limited by these embodiments. The scope of protection of the present invention shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0395] 10: Device that generates a dataset for artificial intelligence training at the interface driver stage 100: Communication unit 200: Storage unit 300: Display unit 400: Voice output unit 500: PLC MAP Standardization Department 600: SVID Standardization Department 700: Semantic group definition 800: Dataset schema definition 900: Dataset Generative Interface Driver 1000: Data Integrity Verification Unit 1100: Key Tag Condition Evaluation Unit 1200: Threshold comparison unit 1300: Record generation unit 1400: Serialization Unit 1500: Integrated Control Unit

Claims

Claim 1 A communication unit that collects physical address-based data from a PLC within the equipment; a PLC MAP standardization unit that converts the collected physical address-based data into a pre-set multi-layered PLC Map to define logical tags; an SVID standardization unit that maps the logical tags to Status Variable IDs (SVID) of the SECS / GEM (Semiconductor Equipment Communication Standard / Generic Equipment Model) standard; a semantic group definition unit that defines semantic groups based on the defined logical tags or the mapped Status Variable IDs (SVID); and a dataset schema definition unit that defines the structure of a dataset for artificial intelligence learning to be generated. A dataset generation interface driver that, in accordance with a user request, among a plurality of physical address-based data stored in a storage unit, generates a data snapshot synchronized by semantic group units for one or more data related to the user request according to the structure of the AI ​​learning dataset to be generated defined above, and converts it into an AI learning dataset structure; a data integrity verification unit that calculates an integrity score by performing pre-set numerical validity verification, state transition-based verification, cross-tag consistency verification, and process rule and physical constraint-based verification on the generated AI learning dataset; and a key tag condition evaluation unit that checks whether to generate a record for the generated AI learning dataset based on the record generation conditions for key tags defined in the dataset schema definition unit.A device for generating an artificial intelligence learning dataset in an interface driver step, comprising a threshold comparison unit that determines whether the integrity score calculated above is greater than or equal to a preset threshold and whether the generated artificial intelligence learning dataset satisfies the record generation condition, wherein the threshold comparison unit requests a record generation unit to generate a record when, as a result of the determination, the integrity score calculated above is greater than or equal to a preset threshold and the generated artificial intelligence learning dataset satisfies the record generation condition, and the record generation unit stores the artificial intelligence learning dataset in the storage unit in accordance with the record generation request from the threshold comparison unit. Claim 2 An apparatus for generating an artificial intelligence learning dataset in an interface driver stage, characterized in that, in claim 1, the preset multi-layer structure includes a physical address layer, a logical tag layer, an SVID standardization layer, a semantic group layer, and an artificial intelligence feature definition layer. Claim 3 A device for generating an artificial intelligence learning dataset at the interface driver stage, characterized in that, in accordance with claim 1, when a project file is uploaded according to user input, it further includes an integrated control unit configured to maintain logical tags and schema structures even if the physical address related to the PLC is changed by linking with the PLC MAP standardization unit and the dataset schema definition unit. Claim 4 A step of collecting physical address-based data from a PLC within the facility by a communication unit; a step of defining a logical tag by converting the collected physical address-based data into a pre-set multi-layered PLC Map by a PLC MAP standardization unit; a step of mapping the logical tag to a Status Variable ID (SVID) of the SECS / GEM standard by an SVID standardization unit; a step of defining a semantic group based on the defined logical tag or the mapped Status Variable ID (SVID) by a semantic group definition unit; and a step of defining the structure of an artificial intelligence learning dataset to be generated by a dataset schema definition unit. A step of converting one or more data related to the user request into an AI learning dataset structure by means of a dataset generation interface driver, according to a user request, by generating a data snapshot synchronized by semantic group units among a plurality of physical address-based data stored in a storage unit, according to the structure of the AI ​​learning dataset to be generated defined above; a step of calculating an integrity score by means of a data integrity verification unit, by performing a pre-set numerical validity verification, state transition-based verification, cross-tag consistency verification, and process rule and physical constraint-based verification on the generated AI learning dataset; a step of determining whether to generate a record for the generated AI learning dataset based on the record generation conditions for key tags defined in the dataset schema definition unit by means of a key tag condition evaluation unit; a step of determining, respectively by means of a threshold comparison unit, whether the calculated integrity score is greater than or equal to a pre-set threshold and whether the generated AI learning dataset satisfies the record generation conditions; and, as a result of the determination, when the calculated integrity score is greater than or equal to a pre-set threshold and the generated AI learning dataset satisfies the record generation conditions, the threshold comparison unit determines that the record generation is performed by the record generation unit. Requesting step;A method for generating an artificial intelligence learning dataset in an interface driver step, comprising the step of storing the artificial intelligence learning dataset in the storage unit in accordance with a record generation request in the threshold comparison unit by the record generation unit. Claim 5 A method for generating an artificial intelligence learning dataset in an interface driver step, wherein, in claim 4, the artificial intelligence learning dataset to be generated is defined as a table structure, converted into a tensor structure or a vector structure according to the learning model input format, and includes a timestamp, equipment ID, lot ID, process step, SVID group, derived feature value, logical tag value, label value, version ID, raw value, and derived variable. Claim 6 delete Claim 7 A method for generating an artificial intelligence learning dataset in an interface driver step, characterized in that, in claim 4, when one artificial intelligence learning dataset is selected by user selection from a list corresponding to artificial intelligence learning datasets stored in the storage unit and a dataset format is selected, the serialization unit converts the selected artificial intelligence learning dataset into the selected dataset format; the integrated control unit transmits the artificial intelligence learning dataset converted into the dataset format to an artificial intelligence server through the communication unit; the integrated control unit receives feedback through the communication unit corresponding to the artificial intelligence learning dataset converted into the corresponding dataset format transmitted from the artificial intelligence server in response to the transmission; and the integrated control unit performs an SVID update function and a control function for the facility based on the received feedback.

Images