WO2021107180A1

WO2021107180A1 - Manufacturing big data-based machine learning apparatus and method

Info

Publication number: WO2021107180A1
Application number: PCT/KR2019/016445
Authority: WO
Inventors: 박덕근; 김유진; 권장환; 신윤수; 백우진; 윤종필; 강인식
Original assignee: 위즈코어 주식회사; 에스케이텔레콤 주식회사
Priority date: 2019-11-27
Filing date: 2019-11-27
Publication date: 2021-06-03

Abstract

The manufacturing big data-based machine learning apparatus and method according to the present invention provide a vibration data-based method for preprocessing and anomaly detection, an apparatus for manufacturing big data processing, and manufacturing big data-based machine learning apparatus and method.

Description

Machine learning device and method based on manufacturing big data

The present invention relates to a machine learning apparatus and method based on manufacturing big data.

According to the 『2018 Data Industry Status Survey』 announced by the Korea Data Industry Promotion Agency, the adoption rate of big data in manufacturing companies is 12.6%. 61.8% of manufacturing companies are not even trying to discuss introduction. The main reason is the lack of information on manpower (39.9%) and the market (35.8%). Lack of information on products or services (35.8%) is also one of the reasons for the delay in the introduction of big data. A manufacturing-related big data processing method applicable to this is required.

In addition, another object to be solved by the present invention is to provide a machine learning apparatus and method based on manufacturing big data capable of preprocessing according to real-time, variability, multidimensional, structured/unstructured, and large-capacity data characteristics.

In addition, another problem to be solved by the present invention is to provide a machine learning apparatus and method based on manufacturing big data that can process streaming data based on high-speed parallel processing using machine learning.

Another problem to be solved by the present invention is to provide a machine learning apparatus and method based on manufacturing big data that can configure and combine data sets multiplexed based on attribute (KEY).

Another problem to be solved by the present invention is to provide a machine learning apparatus and method based on manufacturing big data having a dataset history and partition management function for analysis support using machine learning.

A manufacturing big data-based machine learning apparatus and method according to an embodiment of the present invention provides a vibration data-based pre-processing and anomaly detection method, a manufacturing big data processing apparatus, and a manufacturing big data-based machine learning apparatus and method.

The vibration data-based pre-processing and abnormality detection method according to an embodiment of the present invention includes the steps of collecting vibration-related data in a manufacturing process, pre-processing the collected data, and applying the pre-processed data to a model to perform abnormal detection. input, and detecting an abnormality.

The manufacturing big data processing apparatus according to an embodiment of the present invention includes the steps of collecting historical data related to an event, mapping the collected historical data to be tagged, clustering the mapped historical data, and collecting the clustered historical data. A memory comprising instructions configured to perform the step of visualizing and processors operatively coupled to the memory.

Manufacturing big data-based machine learning apparatus and method according to an embodiment of the present invention includes receiving manufacturing-related data, based on the received manufacturing-related data, using a manufacturing risk prediction model configured to predict anomalies , predicting the anomaly in the manufacturing step, and providing the predicted anomaly in the manufacturing step.

In addition, the present invention has an effect of providing a machine learning apparatus and method based on manufacturing big data capable of preprocessing according to real-time, variability, multidimensional, structured/unstructured, and large-capacity data characteristics using machine learning.

In addition, the present invention has the effect of providing a machine learning apparatus and method based on manufacturing big data that can process high-speed parallel processing-based streaming data using machine learning.

The present invention has the effect of providing a machine learning apparatus and method based on manufacturing big data that can configure and combine data sets multiplexed based on attribute (KEY).

The present invention has the effect of providing a machine learning apparatus and method based on manufacturing big data having a dataset history and partition management function for analysis support using machine learning.

1 is a block diagram for explaining a machine learning apparatus and method based on manufacturing big data according to an embodiment of the present invention.

2 is a schematic diagram for explaining a machine learning apparatus and method based on manufacturing big data according to an embodiment of the present invention.

3 to 8 are schematic diagrams for explaining specific details of a vibration data-based preprocessing and anomaly detection apparatus according to an embodiment of the present invention.

9 to 15 are schematic diagrams for explaining specific details of a manufacturing big data processing apparatus according to an embodiment of the present invention.

16 to 17 are schematic diagrams for explaining specific details of a manufacturing big data-based machine learning apparatus and method according to an embodiment of the present invention.

Advantages and features of the present invention and methods of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be embodied in various different forms, and only these embodiments allow the disclosure of the present invention to be complete, and common knowledge in the art to which the present invention pertains It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims.

Although the first, second, etc. are used to describe various elements, these elements are not limited by these terms, of course. These terms are only used to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may be the second component within the spirit of the present invention.

Like reference numerals refer to like elements throughout.

Each feature of the various embodiments of the present invention may be partially or wholly combined or combined with each other, and as those skilled in the art will fully understand, technically various interlocking and driving are possible, and each embodiment may be implemented independently of each other, and It may be possible to implement together in a related relationship.

Hereinafter, various embodiments of the present invention will be described in detail with reference to the accompanying drawings.

The device may include a communication unit 110 , a user input unit 120 , an output unit 130 , a memory 140 , an interface unit 150 , a control unit 160 , and a power supply unit 170 . Since the components shown in FIG. 1 are not essential, an apparatus having more or fewer components may be implemented.

Hereinafter, the components will be described in turn.

The communication unit 110 may include one or more modules that enable wired/wireless communication between the device and the network in which the device is located. The communication unit 110 transmits/receives a signal to and from at least one of an external device and a server on a communication network such as the Internet. The signal may include various types of data. The communication unit 110 may receive video data from various devices.

The user input unit 120 generates input data for the user to control the operation of the device. The user input unit 120 may include a keypad, a dome switch, a touch pad (static pressure/capacitance), a jog wheel, a jog switch, and the like.

The output unit 130 is for generating an output related to visual, auditory or tactile sense, and may include a display unit 131 , a sound output module 132 , and the like.

The display unit 131 displays (outputs) information processed by the device. For example, the device displays a user interface (UI) or graphic user interface (GUI) related to the system.

The display unit 131 may include a liquid crystal display (LCD), a thin film transistor-liquid crystal display (TFT LCD), an organic light-emitting diode (OLED), and a flexible display (Flexible Display). display) and at least one of a three-dimensional display (3D display). Through the display unit 131 , a moving picture may be displayed, frames of the moving picture may be displayed, and an interface for selecting frames and 3D rendering may be displayed.

The sound output module 132 may output audio data received from the communication unit 110 or stored in the memory 160 . The sound output module 132 also outputs a sound signal related to a function performed by the device.

The memory unit 140 may store a program for processing and control of the controller 160 , and may perform a function for temporarily storing input/output data. The memory 140 may include a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (eg, SD or XD memory), and a RAM. (Random Access Memory, RAM), SRAM (Static Random Access Memory), ROM (Read-Only Memory, ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, magnetic It may include at least one type of storage medium among a disk and an optical disk. The device may operate in relation to a web storage that performs a storage function of the memory 160 on the Internet.

The interface unit 150 serves as a passage with all external devices connected to the device. The interface unit 150 receives data from an external device, receives power and transmits it to each component inside the device, or allows data inside the device to be transmitted to an external device. For example, wired/wireless headset ports, external charger ports, wired/wireless data ports, memory card ports, ports for connecting devices equipped with identification modules, audio input/output (I/O) ports, A video input/output (I/O) port, an earphone port, etc. may be included in the interface unit 150 .

A controller 160 typically controls the overall operation of the device. For example, it performs processing of data or related control and processing for displaying processed data. The controller 160 may include a graphic module 161 for parallel data processing. The graphic module 161 may be implemented within the control unit 160 or may be implemented separately from the control unit 160 .

The power supply unit 170 receives external power and internal power under the control of the control unit 160 to supply power necessary for the operation of each component.

Various embodiments described herein may be implemented in a computer-readable recording medium using, for example, software, hardware, or a combination thereof.

According to the hardware implementation, the embodiments described herein include application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), It may be implemented using at least one of processors, controllers, micro-controllers, microprocessors, and other electrical units for performing functions. The described embodiments may be implemented by the controller 160 itself.

According to the software implementation, embodiments such as the procedures and functions described in this specification may be implemented as separate software modules. Each of the software modules may perform one or more functions and operations described herein. The software code may be implemented as a software application written in a suitable programming language. The software code may be stored in the memory 160 and executed by the controller 160 .

Time-series, variable, multidimensional, and large amounts of data are stored in the original DB, and continuous data sources may have some missing values. The preprocessing engine automates preprocessing through preprocessing parallel execution threads. For continuous data, missing values may be supplemented through interpolation, for example, and missing values may be added using statistical information. Pattern-type data can be supplemented according to the pattern. Unstructured data can be converted into structured data and preprocessed by pattern complementation, interpolation complementation, or statistical addition.

Next, an access query optimization operation to speed up a large data query may be performed. In addition, for unstructured and patterned data, refinement can be performed for optimization search query. High-speed or sophisticated data can be set to be accessible as a service API. In addition, it may be set to be searched by metadata, managed to allow modification of metadata, and monitoring of storage management performance may be possible.

Hereinafter, specific details of the vibration data-based preprocessing and anomaly detection apparatus according to an embodiment of the present invention will be described with reference to FIGS. 3 to 8 .

Referring to FIG. 3 , an apparatus capable of collecting vibration-related data in a chamber using plasma and detecting abnormality by vibration will be described.

Referring to FIG. 4 , the core of the present system is an algorithm for detecting an important change (event) occurring in the process and an algorithm for detecting an abnormal situation in the process based on a separate event. The process event detection module detects the start and end times of the process and plays a role in classifying the data of the actual process period. This module is composed of the following detailed modules. The sensor data input manager is responsible for fetching continuously incoming sensor data from the database. The data preprocessor determines whether there is noise in the data that does not affect the process due to the unique characteristics of the facility or special environment settings, removes the noise if necessary, and transforms the data into a form suitable for analysis (Noise Smoothing Operator) do. The event detection manager is responsible for detecting the actual start and end of the process. To this end, it analyzes the variability and directionality of data (Changing State Analyzer) and detects important changes in the process to determine the start and end time of the process (Event Detector). The event detection algorithm that changes from non-process to process state (Start Event) or from process to non-process state (End Event) proceeds in the following order.

Step 1: Ratio X±3σ using the mean (X) and standard deviation (σ) based on the data values obtained for a certain period of time under the assumption that the non-process state is maintained for a certain period of time after the process starts to detect the event Set in the range of the process state.

Step 2: Define 6 types of general types in which the direction of data is changed for the range of the set non-process state. In the non-process state, the direction is changed in an increasing direction (1). In an increasing direction, it is changed in a decreasing direction (2), or it is changed to a non-process state (3). In the decreasing direction, it is changed in the increasing direction (4), or it is changed to a non-process state (5). And there are cases where the same direction is maintained (6).

In case of (4), it means an event that starts at the same time as the process ends. Step 4: In order to distinguish the process start and end of the selected sensor data based on

steps

2 and 3, a core sensor with clear characteristics of process and non-process states is used as an auxiliary sensor. That is, if the minimum threshold determined as a process event is specified and the selected sensor enters the start event state and the number of auxiliary sensors in the event state exceeds the threshold, it is determined as the start of the actual process and the same method is used for the end event.

The data division manager divides the actual process data (Event da-ta) based on the start and end times of the process determined by the event detection manager and stores it in the storage (Segmented Data Storage) (Segmentation Operator).

Using the actual process part data stored in the segmented data storage, the similarity with the reference pattern stored in the reference pattern storage is measured to determine whether the current process is proceeding normally, and abnormal This is a module that notifies the operator of the current process status when a suspected result is obtained. The process parameter data extracted from the inline facility has a characteristic of having a length dependent on the amount of input LCD panel. Therefore, when measuring the similarity of two data with different input amounts, preprocessing is required to adjust the length to be similar to each other. In the case of processing multiple LCD panels, most of the data extracted during the operation of the equipment has an internal cycle or is processed within a certain boundary.

The preprocessing of inline facility data is performed for the purpose of matching the lengths of the two data for pattern matching as closely as possible. For this, a preprocessing rule of calculating the length ratio between the separated actual process data and the reference pattern and duplicating the data was used. The preprocessing rule replicates the short-side data by a factor close to the long-side length so that the short-length data approximates the long-side data (Rule 1), or duplicates the long-side data by 2 times and the short-side data by 3 times ( Rule 2), using the original two data as it is (rule 3) is used. In addition to these three rules, a closer result can be obtained by varying the replication magnification, but considering that the performance improvement effect is not as great as the similarity measurement speed and diversity of pattern matching, other rules are not considered. Rule 1, Rule 2, and Rule 3 are in principle used to make the length ratio of the two data closest to each other.

The process anomaly detection manager extracts the reference pattern of the corresponding sensor data based on the sensor information to be analyzed (Pattern Selector) and measures the similarity through pattern matching with the process data (Similar-ity Measuring Operator).

Referring to FIG. 5 , a watt-hour meter, a temperature sensor, a pressure gauge, a rectifier current monitoring unit, and an ACDC converter and a communication module connected to each are shown. Data can be collected from the communication modules to the device through the hub and stored in the DB.

Referring to FIG. 6 , the device may perform real-time parameter monitoring, abnormal parameter detection, recipe management, recipe quality management, real-time equipment monitoring, and equipment performance analysis, and each report may be provided.

Referring to FIG. 7 , the device may be divided into Yield Management System, Fault Detection & Classification, and Equip Predictive Maintenance parts, and each part may use a pre-trained deep learning model or a deep reinforcement learning model.

Referring to FIG. 8 , an example of detecting an abnormal signal through time series analysis is illustrated, and an example of predicting Perriodic and Major Trend information by deep learning is illustrated.

Hereinafter, specific details of the manufacturing big data processing apparatus according to an embodiment of the present invention will be described with reference to FIGS. 9 to 15 .

Referring to FIG. 9 , a method of utilizing data assets for smart factory management through optimization is described. Past data, present data and future data can be used to predict the future situation. Through this, the production and manufacturing status can be quickly and accurately grasped.

Referring to FIG. 10 , factories and respective lines in different regions are connected through an operation network, and an operation side and a business office are connected.

Referring to FIG. 11 , a flow diagram of a connection relationship from an infrastructure to a user for connection, analysis, and insight provision in devices, machines, and systems is shown.

12 , a manufacturing big data processing method is described that collects data from sources including plants, factories, and sites, performs hybrid analytics (using PlantPulse, for example), and makes them accessible to users with applications (eg, using PlantPulse). .

Referring to FIG. 13 , a process in which RAW data is collected as events or points, the collected data is mapped to be tagged, and the tagged data is analyzed through filtering, clustered, and visualized will be described.

Referring to FIG. 14 , a process in which the data of FIG. 7 is collected, tagged, and processed will be described in more detail.

Referring to FIG. 15 , an input event is shown, and an event output through EQL is described. In addition, as examples of analysis utilization, filtering, database lookup, event pattern matching, and the like are described.

Hereinafter, a manufacturing big data-based machine learning apparatus and methods according to an embodiment of the present invention will be described with reference to FIGS. 16 to 17 .

More specifically, the manufacturing risk prediction model configured to predict anomalies of the present invention may be configured to include one hidden layer composed of 64 nodes. Furthermore, in the predictive model, a learning rate value, which may be a parameter for finding a weight that minimizes a prediction error in predictive learning, may be set to 0.0009. Also, a momentum value, which is a parameter value for minimizing prediction error and increasing the learning rate, may be set to 0.9. In addition, the predictive model of the present invention can be configured to use 'rmsprop' as an optimization function to update parameters in learning, and 'as a function to determine the intensity of the input value of various clinical data transmitted to the output value' It can be configured to use the relu' function.

However, the type of prediction model of the present invention is not limited thereto. For example, the prediction model may include a Deep Neural Network (DNN), a CNN Convolutional Neural Network (DNN), a Recurrent Neural Network (RNN), a Deep Convolutional Neural Network (DCNN), a Restricted Boltzmann Machine (RBM), and a Deep Belief Network (DBN). , it may be a single shot detector (SSD) model or a predictive model based on U-net.

Referring to FIG. 16 , the predictive model of the present invention may be a predictive model based on the MLP algorithm, which is a multi-layered artificial neural network. More specifically, the predictive model of the present invention may be a multi-layered prediction model in which an output layer predicting anomalies to which manufacturing-related data is input and one hidden layer exist between these layers.

Referring to FIG. 17 , the evaluation of the predictive model may be performed based on the LRP algorithm or more various algorithms without being limited thereto. For example, data having relevance to predicting anomalies may be determined based on a Randomized Decision forest algorithm or a Penalized Logistic Regression algorithm.

Although the embodiments of the present invention have been described in more detail with reference to the accompanying drawings, the present invention is not necessarily limited to these embodiments, and various modifications may be made within the scope without departing from the technical spirit of the present invention. . Therefore, the embodiments disclosed in the present invention are not intended to limit the technical spirit of the present invention, but to explain, and the scope of the technical spirit of the present invention is not limited by these embodiments. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. The protection scope of the present invention should be construed by the following claims, and all technical ideas within the equivalent range should be construed as being included in the scope of the present invention.

[National R&D project supporting this invention]

[Project identification number] 1711093793

[Name of Ministry] Ministry of Science and ICT

[Research management agency] Information and Communication Planning and Evaluation Institute

[Research project name] High-performance platform/SW technology

[Research project name] 5G-based production/logistics management service and cloud-oriented manufacturing-specialized ML platform development

[Contribution rate] 1/1

[Organizer] SK Telecom Co., Ltd.

[Research period] 2019.01.01 ~ 2019.12.31

Claims

collecting data related to vibration in the manufacturing process;

preprocessing the collected data; and

Pre-processing and anomaly detection method based on vibration data, comprising the step of inputting the preprocessed data into a model to perform abnormality detection, and detecting anomaly.
collecting historical data related to the event;

mapping the collected historical data to be tagged;

clustering the mapped historical data, and

a memory comprising instructions configured to perform the step of visualizing clustered historical data, and

and processors operatively coupled with the memory.
receiving data associated with manufacturing;

predicting an anomaly in a manufacturing step using a manufacturing risk prediction model configured to predict the anomaly based on the received manufacturing-related data; and

A machine learning method based on manufacturing big data, comprising the step of providing anomalies in a predicted manufacturing step.