JP7169706B2 - Algorithms and methods for detecting machine error data based on machine learning techniques - Google Patents

Description

The present invention relates to algorithms and methods for detecting machine error data based on machine learning techniques.

The rapid spread of the Internet and the development of technology for connecting all things based on the Internet have led to the rise of the fourth industrial revolution. One of the various technical fields that are leading the 4th Industrial Revolution is technology related to smart factories.

A smart factory is a holistic technology that connects factory machines and facilities for easy management. One of the various technologies that make up such a smart factory is the technology for managing machines. In other words, it monitors whether or not the machine is operating normally, and predicts and manages the occurrence of errors in advance.

Conventionally, in order to predict machine malfunction, a threshold defined in the form of an absolute value that is not related to time is set, machine operation data is collected at any time, and operation data exceeding the threshold is generated. is determined to have caused a specific error. Conventional thresholds relate to machine set tolerances for production, in other words, combinations of upper and lower limits of operation that do not result in defective products being produced. Conventional thresholds are fixed values directly set by factory workers (absolute values that are fixed irrespective of time in each section set by workers). Therefore, if the worker wants to reduce the defect rate by arbitrarily degrading the quality of the product, the upper and lower limits of the threshold should be set wider. In the opposite case, the upper and lower limits of the threshold should be set narrower. do it. Also, a threshold value is set for a time interval during which a product is manufactured during a period corresponding to one cycle in which the machine operates, and a separate threshold value is not set for a time interval that is not related to product manufacturing. Therefore, although it is possible to determine the presence or absence of defective products, it is difficult to accurately determine the presence or absence of machine errors.

The present invention is intended to solve the problems of the conventional technology described above, and automatically extracts chronological threshold data from machine operation data collected in real time based on machine learning techniques, thereby enabling The purpose is to enable precise prediction and detection of machine error data that could not be comprehended because the threshold was manually set by the operator.

Another object of the present invention is to provide error data that exceeds chronological threshold data so that it can be easily confirmed at the operator's terminal.

According to one embodiment of the present invention, a method for detecting error data in a machine based on machine learning techniques includes the steps of (a) collecting time-series operational data of at least one machine; (c) time-series standard data for the mapped set of motion data based on machine learning; (d) judging an error event if the operation data collected in real time exceeds the threshold data, and providing information about the error event to the worker terminal; and the step of

A server for detecting machine error data based on the machine learning technique according to another embodiment of the present invention includes a memory storing a program for detecting machine error data based on the machine learning technique, and executing the program. and a processor for executing the program, the processor collects time-series operational data of at least one machine, divides the operational data into predetermined time intervals, and performs the divided operations Data is mapped so that it overlaps on the same time domain, and time-series standard data for the set of mapped motion data is derived based on machine learning to generate time-series threshold data and collect it in real time. If the received motion data exceeds the threshold data, it is determined as an error event, and information about the error event is provided to the operator terminal.

The present invention automatically detects chronological threshold data in the server and compares the operation data in all time domains based on it, so it is possible to detect machines that could not be found when setting the conventional threshold (absolute value). defects and product defects can be precisely detected.

As a result, the present invention can check the presence or absence of abnormalities in products and quality more elaborately than in the conventional art, so precise predictive maintenance is possible.

In addition, since the thresholds are generated precisely in chronological order for all time domains, the accuracy of defect/malfunction detection of machines and products is extremely high, which allows the process capability index (Cp) of the machine to be calculated. It can be greatly improved and accurate process capability index detection is possible.

Further, by providing the user interface of the worker terminal so that the worker can easily check the state of each machine, convenience for the worker can be improved.

1 is a structural diagram of a system according to an embodiment of the present invention; FIG. 1 is a block diagram of the structure of a sensor assembly according to one embodiment of the present invention; FIG. 3 is a graph of three representative types of operational data measured from a machine; FIG. 3 is a block diagram of the structure of a server according to one embodiment of the present invention; FIG. 4 is a graph of preprocessed motion data according to one embodiment of the present invention; FIG. FIG. 10 is a graph obtained by dividing motion data in units of 60 seconds and then mapping each divided area so as to overlap in units of 60 seconds according to an embodiment of the present invention; FIG. 4 is a graph of standard data detected based on machine learning from motion data collected according to an embodiment of the present invention; 5 is a graph of time-series threshold data according to an embodiment of the present invention; FIG. 5 is a graph of arbitrary motion data input to validate threshold data, according to one embodiment of the present invention. FIG. FIG. 4 is a graph when error data occurs according to an embodiment of the present invention; FIG. It is an example of a user interface according to one embodiment of the present invention. FIG. 5 is a graph for comparing threshold data according to one embodiment of the present invention and the prior art; FIG. 1 is a flowchart illustrating a method for detecting machine error data based on machine learning, according to one embodiment of the present invention;

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement the present invention. However, the present invention may be embodied in various different forms and should not be construed as limited to the embodiments set forth herein. In addition, in the drawings, in order to clearly explain the present invention, parts that are not related to the description are omitted, and like reference numerals are attached to like parts throughout the specification.

In the entire specification, when a part is "connected" to another part, it means not only "directly connected" but also "electrically Including cases where they are “connected”. In addition, when a part "includes" a component, it means that it can further include other components, rather than excluding other components, unless specifically stated to the contrary.

In this specification, the term "unit" includes a unit implemented by hardware, a unit implemented by software, and a unit implemented using both. Also, one unit may be implemented using two or more pieces of hardware, and two or more units may be implemented by one piece of hardware. On the other hand, "section" is not meant to be limited to software or hardware, and "section" may be configured to reside on an addressable storage medium and reproduce one or more processors. It may be configured as follows. Thus, by way of example, "part" includes components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, program code Includes segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays and variables. Functionality provided in components and "sections" may be combined into fewer components and "sections" or further separated into additional components and "sections". Moreover, the components and "parts" may be embodied to play one or more CPUs in a device or secure multimedia card.

A "terminal" referred to below may be embodied in a computer or portable terminal capable of connecting to a server or other terminal via a network. Here, the computer is, for example, a notebook computer equipped with a web browser (WEB Browser), a desktop (desktop), a laptop (laptop), a VR HMD (e.g., HTC VIVE, Oculus Rift, GearVR, DayDream, PSVR, etc.) etc. may be included. Here, VR HMDs are for PC (e.g. HTC VIVE, Oculus Rift, FOVE, Deepon, etc.), mobile (e.g. GearVR, DayDream, Storm Mirror, Google Cardboard, etc.), and console (PSVR). It includes all independently implemented Stand Alone models (eg, Deepon, PICO, etc.). Portable terminals are, for example, wireless communication devices that ensure portability and mobility, and include not only smart phones, tablet PCs, wearable devices, but also Bluetooth (BLE, Bluetooth Low Energy), NFC, and RFID. , Ultrasonic, Infrared, WiFi, LiFi, etc. equipped with communication modules. In addition, a 'network' means a connection structure capable of exchanging information between nodes such as a terminal and a server, and includes a local area network (LAN) and a wide area network (WAN). : Wide Area Network), the Internet (WWW: World Wide Web), wired and wireless data communication networks, telephone networks, wired and wireless television communication networks, and the like. Examples of wireless data communication networks include 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, Bluetooth communication, infrared Communication, Ultrasonic Communication, Visible Light Communication (VLC), LiFi, etc., but not limited thereto.

In the following, "operating data" means data that can directly or indirectly indicate the operation of the machine, such as information such as machine temperature, humidity, pressure, power, etc. You can

An "object" means a part produced in a factory or a lower component of the part, and means an object manufactured/produced by one machine.

Also, "error data" means data relating to at least one of machine errors and machine-produced object defects.

Referring to FIG. 1, a system according to one embodiment of the present invention includes a sensor assembly 100 installed near a machine 10 in a factory, a worker terminal 150, a server 200, and an administrator terminal 300. . Here, the worker terminal 150 means a terminal assigned to a worker who is in charge of any one process among line equipment in a factory, and the manager terminal 300 is assigned to a line manager or a factory manager. It can also mean a terminal. The worker terminal 150 and the manager terminal 300 may be collectively referred to as a worker terminal.

A system according to one embodiment of the present invention is a system capable of providing smart factory services. Such a smart factory service monitors the operation status of the machine equipment 10 in the factory in real time, and immediately notifies the manager when there is a high possibility of malfunction or when a malfunction occurs. It can provide administrators with efficiency and convenience in machine 10 management. In particular, by providing Internet Of Things (IOT) technology-based services through the sensor assembly 100, the factory manager can eliminate the inconvenience of inquiring the machines 10 one by one to check whether there is a problem. By performing machine learning on the values measured in , and automatically setting threshold data in the server, it is possible to precisely determine whether there is an abnormality in the machine or in the product.

The sensor assembly 100 is an IOT integrated module terminal configured with at least one sensor. The sensor assembly 100 is installed in the vicinity of the machine 10 in the factory and may be configured to be attached to any one surface of the machine 10 . Sensor assembly 100 includes sensors for measuring operational data of machine 10 and sensors for transmitting it to server 200 .

Specifically, referring to FIG. 2 , sensor assembly 100 includes sensor controller 110 , communication module 120 , measurement sensor 130 and connector 140 .

Each of the sensor controllers or measurement sensors 110-130 may be embodied in a physically independent form. In other words, as shown in FIG. 1, each sensor may be formed in a hexahedral shape having similar or identical physical specifications, making it very easy to replace each module as required. For example, if any one of the sensor controller 110, the communication module 120, or the measurement sensor 130 malfunctions or the application needs to be changed, the problem can be easily solved by replacing the sensor. can. Moreover, the sensor assembly 100 may further include a stand. The pedestal functions to support both the sensor controller 110 and the communication module 120 . The pedestal has an area that can cover the area of the sensor controller 110 and the communication module 120, and a partition is formed in the edge area to fix the sensor controller 110 and the communication module 120 so that they do not move to the outside. can be At this time, the sensor controller 110 and the communication module 120 may be arranged in a stacked form on the stand.

The sensor controller 110 receives the electrical signal value (current or voltage value) measured by the measurement sensor 130 from the measurement sensor 130 and notifies the communication module 120 to transmit the electrical signal value to the server 200 for collection. At this time, the sensor controller 110 is connected with at least one measurement sensor 130 . Even if the previously connected measurement sensor 130 is replaced with another type of measurement sensor 130 due to replacement or addition of the measurement sensor 130, the sensor controller 110 receives the electrical signal value of the currently connected measurement sensor 130. can be recognized by

For example, various sensors such as temperature sensors, pressure sensors, humidity sensors, current/voltage sensors, and power sensors may be coupled to the sensor controller 110 . Further, when the firmware is installed in the server 200, the sensor controller 110 simply transmits the electric signal value (that is, the A/D signal value: the signal converted from analog to digital) of the measurement sensor 130 to the server 200. If the sensor controller 110 has downloaded and installed the firmware for all the sensors, the sensor controller 110 will be able to recognize the signals of the sensors, even if the sensors are different and incompatible with each other. can be done.

The sensor controller 110 converts the operational data received from the measurement sensor 130 into a standardized digital signal and transmits it to the communication module 120 . For example, the signals transmitted from the temperature sensor and the pressure sensor to the sensor controller 110 are electrical signals in different formats. If this is transmitted to the server 200 as it is, there is a possibility that the server 200 cannot accurately recognize what kind of information is included. Therefore, the sensor controller 110 may play a role of converting the analog log signal or digital signal into a standardized digital signal into a form recognizable by the server 200 .

The communication module 120 communicates information between the sensor controller 110 and the server 200 or user terminal 300 . The electrical signal values sent by the communication module 120 to the server 200 can be recognized as operational data by firmware stored in the server 200 . In other words, it is sent to the server 200 as a simple current value or voltage value, but since there is firmware in the server 200, it can be recognized as a value related to operational data such as temperature, pressure, and humidity.

Measurement sensor 130 is a sensor that measures operational data of machine 10 . For example, the measurement sensor 130 may be a sensor that measures any one of temperature, pressure, humidity, voltage, power, and vibration. This is just one example and may include sensors that measure a variety of other operational data.

Sensor controller 110 , communication module 120 and measurement sensor 130 may be coupled and secured to each other via connector 140 . The connector 140 may be embodied in the form of wires or wires. Alternatively, connector 140 may be embodied in the form of a plurality of pins formed in one area of each sensor. When embodied in the form of pins, the sensors may be connected to each other by arranging and connecting the connectors 140 formed on the respective sensors so as to be engaged with each other. In addition, when the connectors 140 are provided in a form that can be fixedly coupled to each other, it is possible to provide not only the connection between the connectors 140 but also the effect of fixing the positions of the sensors.

The server 200 can receive electrical signal values relating to operation data of the machines 10 from the sensor assemblies 100 provided for each machine 10, and can recognize which operation data are related based on the firmware. In addition, by automatically generating time-series threshold data for the machine based on operation data collected by analysis methods based on big data analysis and machine learning, workers do not have to manually enter threshold data corresponding to absolute values. Since it is threshold data based on machine learning, it is possible to set an accurate threshold, and it is possible to accurately determine the presence or absence of machine errors and product defects.

Information about the operational data of the servers 200,100 may be provided to the worker terminals 150,150 and the manager terminals 300,300.

The worker terminals 150, 150 are terminals installed in the machine 10 or arranged near the machine 10, and directly monitor and check the status of the machine 10 that the worker in charge of one process is in charge of. Where possible, real-time operational data for the machine 10 may be displayed.

An application that provides smart factory services may be installed on the administrator terminals 300 , 300 . The application receives information from the servers 200 and 200, processes it into a form that is easy for the user to understand, provides the user with information on the operation status of the machines 10 and the equipment 10, and provides operational data for the entire machine 10 in the factory at a glance. It may be displayed so that it can be queried.

Hereinafter, the error detection method of the machine 10 based on the machine learning of the server 200 according to the present invention will be specifically described.

Typical types of operation data of the machine 10 are summarized into about three types as shown in FIG. That is, motion data that has different amplitudes depending on time and repeats at a specific cycle like 1, motion data that has the same amplitude and cycle depending on time and changes frequency like 2, and 3. After the machine 10 operates for a certain period of time, it can be summarized as operation data that becomes saturation.

A conventional machine 10 error or product error detection method is a method in which an operator manually inputs a threshold value, and if operation data exceeding the threshold value is found, it is recognized as an error event. At this time, the threshold is an absolute value defined independently of time. Therefore, as in 3, the threshold value for the saturation value is set and the conventional method is appropriate when judging whether or not the operation is normal. If the frequency changes or the frequency changes, the existing method cannot recognize an accurate error event. Therefore, when the conventional method is applied to 1, the absolute value threshold is applied only to certain time interval regions directly linked to product production among all time regions. Therefore, error events in other time interval regions to which the absolute value threshold is not applied cannot be detected.

An embodiment of the present invention, which will be described later, generates time-series threshold data that matches the pattern of the operation data based on a machine learning technique, regardless of the type of operation data of the machine 10. It can be said that the error detection method of the machine 10 is applicable to the machine 10 related to the type of operation data. In the following description, it is assumed that the operating data of the machine 10 are of type 1.

Referring to FIG. 4, the server 200 includes a memory storing a program (or application) that performs a method for detecting error data of the machine 10 based on a machine learning method, and a processor that executes the program. can be Here, the processor may perform various functions by executing programs stored in the memory, and detailed components included in the processor according to each function are data preprocessing unit 210, data analysis unit 220, and so on. Threshold data detector 230 , error detector and judger 240 , and motion data provider 250 .

The data preprocessing unit 210 collects operation data of the machine 10 from the past to the present and performs preprocessing.

Assuming that the operational data of the machine 10 is of type 1 in FIG. 3 described above, the operational data is formed into a pattern each time the machine 10 fabricates an object. That is, if the machine 10 builds one object for 60 seconds, the operational data will have the same or similar amplitudes (values for temperature, pressure, voltage, etc.) every 60 seconds. At this time, operational data collected while the machine 10 is operating to fabricate objects is meaningful data, so data during times when the machine 10 is not operating is eliminated and operating Data may be collected over time and compression may be performed on the collected data. Thereby, the preprocessed motion data may be presented in a graph in the form of FIG.

At this time, event data 410 having a pattern that deviates from the pattern of collected motion data may be detected. Specifically, event data 410 having a pattern that deviates from the average pattern (for example, graph shape) formed by collected motion data may be detected. Event data 410 is shown whose value appears much higher than other operational data.

Such event data 410 are data relating to machine 10 errors or object errors. At this time, based on the work data recorded at the actual work site, the work information at the moment when the event data 410 was detected (that is, information regarding the presence or absence of an error in the machine 10 or the presence or absence of a defect in the object) is retrieved, and the The work information and the event data 410 may be matched and stored in the server 200 . Information stored matching against the event data 410 includes information indicating machine 10 error and object failure, information indicating machine 10 error and object normal, information indicating machine 10 normal and object failure, and information indicating machine 10 normal and object normal. may be any one of the information indicating

The data analysis unit 220 may derive standard data for the motion data by performing a machine learning technique on the preprocessed motion data.

Specifically, the data analysis unit 220 may divide the motion data into predetermined time intervals, and map the divided motion data on a time domain having a length corresponding to the predetermined time intervals. For example, the predetermined time interval may be one period of motion data (eg, 60 seconds). A cycle may be the time it takes machine 10 to manufacture one object. The data analysis unit 220 divides all the collected time-series motion data into units of one cycle, and maps the values constituting the divided motion data on a graph having the length of one cycle. In that case, a graph like FIG. 6 can be derived. That is, all the points forming the divided motion data are mapped on the graph. According to the graph of FIG. 6, it can be seen that the motion data repeats in a particular pattern.

The data analysis unit 220 extracts at least one time-series standard data based on the average value or median value from the set of mapped motion data based on a machine learning method, and extracts the standard data with the highest K index. data may be detected. When the data analysis unit 220 extracts the standard data based on the median value, the most frequently divided motion data among all the divided motion data superimposed on the graph of FIG. 6 is detected as the standard data. can be Here, the K index may be a statistical index, mean value or median value. The closer the K index of a particular graph is to 1, the closer the graph is to the standard value of the operating data. That is, the data analysis unit 220 repeatedly detects standard data from collected motion data, performs machine 10 learning while measuring the K index, and detects standard data with the highest K index. For example, graphs for standard data such as FIG. 7 can be derived.

The threshold data detection unit 230 detects chronological threshold data based on the standard data. Specifically, by performing the machine learning process performed by the data analysis unit 220 on the motion data having a Y-axis value (amplitude value) higher than the standard data among the motion data mapped on the graph of FIG. Upper limit threshold data 422 is detected. Then, the threshold data detection unit 230 performs the same machine learning process on motion data having a lower Y-axis value (amplitude value) than the standard data among the motion data mapped on the graph of FIG. Lower limit threshold data 423 is detected. At this time, a combination of the upper limit threshold data 422 and the lower limit threshold data 423 becomes the threshold data.

Referring to FIG. 8, it can be seen that standard data 421 is mapped between upper limit threshold data 422 and lower limit threshold data 423 . It can be seen that the threshold data is configured to have different values depending on time. That is, since the type of motion data is type 1 in FIG. 3, the threshold data derived based on machine learning is also type 1. FIG. Although FIG. 8 shows that the upper threshold data 422 and the lower threshold data 423 have a difference of +10 or -10 with respect to the standard data, this is only an example, and other difference values are shown. may be configured.

The error detection and determination unit 240 detects an error by comparing real-time collected motion data and threshold data, and determines the type of error.

If the real-time operational data is configured to be distributed between the upper and lower limits of the threshold data as in FIG. 9, the error detection and decision unit 240 may consider the machine 10 and objects to be in normal condition. .

However, if the real-time motion data 425 exceeds either the upper limit or the lower limit in any one time interval as shown in FIG. Also good. The error data 425 of FIG. 10 has the form of a normal data pattern up to about 25 seconds, but has an abnormal pattern exceeding the threshold data in the time interval d between about 25 seconds and about 50 seconds. is doing.

Error detection and determination unit 240 then compares the pattern of error data 425 to the pattern of event data 410 . The error detection and judgment unit 240 judges the information on the presence of an error in the machine 10 and the presence of a defect in the object in the event data 410 having a pattern corresponding to the error data 425 according to the comparison result as the information on the error data 425 . In other words, the event data 410 is divided into the above four types (machine 10 normal and object normal type, machine 10 normal and object defective type, machine 10 error and object normal type, machine 10 error and object error type). It may be determined which of these four types the error data 425 falls into.

The motion data provider 250 may provide information about error events to the worker terminal.

At this time, the user interface provided to the worker terminal may have a form as shown in FIG. Specifically, the user interface displays identification information (eg, WF-11, WF-12, WF-21, WF-22, WF-31, WF-32 in FIG. 11) regarding a plurality of machines 10 included in the worksite. ), and status information of the machines 10 displayed after being matched for each identification information about each machine 10 and the objects manufactured by the machines 10 . That is, status information may be provided in blocks for each machine 10 . As shown in FIG. 11, the current status information of the machine 10 is represented by displaying different colors and brightness for each status value (machine 10 malfunction, machine 10 normal, defective product, normal product). Also good. Also, in an additional embodiment, the user interface may work in the form of a map by presenting a picture of a structure in which a plurality of machines 10 are arranged, and presenting identification and status information about each machine 10 on the picture. The status information for each machine 10 may be provided to the operator so that he/she can easily check it at a glance.

In addition, when an operator inputs (touches or clicks, etc.) identification information relating to any one of the machines 10 on the user interface, real-time collection information relating to a plurality of chronological operation data for the machines 10 is provided. Also good. For example, real-time graphs of machine 10 temperature, humidity, pressure, etc. may be provided.

The user interface may also provide the operator to overlay the real-time operational data graph and the threshold data graph when the real-time graph is magnified.

On the other hand, even after the threshold data is generated once through the above-described machine learning and big data analysis process, motion data is continuously collected and accumulated. It may be updated by performing the process again.

According to FIG. 12, the threshold data set by the operator according to the prior art are defined as absolute values such as USL (Upper Spec Limit)' and LSL (Lower Spec Limit)'. In this case, an error determination is possible only for motion data within the range of about 10 seconds to about 30 seconds. That is, it is possible to determine whether there is an error in the machine 10 or a defect in the product only in the area E2 of FIG. However, according to one embodiment of the present invention, the threshold data consists of upper threshold data 422 and lower threshold data 423 that change in time series. As a result, it is possible to detect the error data in the area E1, which could not be detected by the conventional technique.

Hereinafter, an error detection method for the machine 10 based on machine learning according to one embodiment of the present invention will be specifically described with reference to FIG. Since the following method is performed by the server 200 described above, even if some parts are omitted, they are considered to be replaced with the contents described above.

The server 200 first collects time-series operation data of the machine 10 and performs preprocessing (S110).

The server 200 extracts a plurality of error patterns from the preprocessed motion data (S120). If a pattern deviating from the average pattern indicated by the operation data is found, it is detected as an error pattern, each error pattern is matched with information on the work diary, and the error pattern indicates the presence or absence of an abnormality in the machine 10 or object. to figure out what it represents.

Next, the server 200 detects chronological threshold data from all motion data collected based on machine learning (S130). Standard data with the highest K index is detected, and threshold data is detected based on the standard data.

The server 200 detects error data exceeding the threshold data (S140).

At this time, the server 200 detects an error state by comparing the error data with a plurality of error patterns (S150). That is, the error state is detected by comparing the error data with the already detected event data 410 .

The server 200 provides error status information according to the user interface that has already been set to the worker terminal (S160).

An embodiment of the present invention may be embodied in the form of a recording medium including computer-executable instructions, such as program modules, executed by a computer. Computer-readable media can be any available media that can be accessed by the computer and includes both volatile and non-volatile media, removable and non-separable media. Also, computer-readable media may include any computer storage media. Computer storage media includes volatile and non-volatile, discrete and non-volatile embodied in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Includes all separate media.

Although the methods and systems of the present invention have been described in connection with specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.

The above description of the present invention is for illustrative purposes only, and a person skilled in the art to which the present invention pertains can make other specific examples without changing the spirit or essential features of the present invention. It should be understood that it can be easily transformed into a Therefore, the above-described embodiments are to be considered in all respects as illustrative and not restrictive. For example, each component described in a single form may be implemented in a distributed fashion, and similarly, components described as distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the claims set forth below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalents are considered within the scope of the present invention. should be construed to fall within the scope of

10: Machine 100: Sensor Assembly 150: Worker Terminal 200: Server 300: Manager Terminal

Claims (12)

In a machine error data detection method based on machine learning techniques performed by a server,
(a) collecting operational data over time for at least one machine;
(b) dividing the motion data into predetermined time intervals and mapping the divided motion data so as to overlap on the same time domain;
(c) generating time-series threshold data by deriving time-series standard data for the set of mapped motion data based on machine learning;
(d) judging an error event if the operational data collected in real time exceeds the threshold data, and providing information about the error event to the operator terminal;
Said step (a) comprises:
detecting event data having a pattern that deviates from the average pattern of the collected motion data;
matching and storing information regarding the presence or absence of errors in the machine and the presence or absence of defects in the object manufactured by the machine with respect to the event data;
The step (b) includes
a step of dividing the collected motion data by designating one cycle of the motion data as the predetermined time interval, and mapping the divided motion data on the time domain having a length corresponding to the one cycle; including
The step (d) includes
If error data exceeding the threshold data is detected at any one point in the operation data collected in real time, it is determined whether there is an error in the machine and whether there is a defect in the object based on event data corresponding to the pattern of the error data. including steps
A machine error data detection method based on machine learning techniques, wherein one period of the motion data is the time it takes for the machine to manufacture one object. Said step (a) comprises:
2. The method of detecting error data for a machine based on machine learning techniques of claim 1, comprising collecting motion data as the machine operates to perform a specified object manufacturing process. The information stored matching the event data is any one of information indicating a machine error and an object failure, information indicating a machine error and an object normality, and information indicating a machine normality and an object failure. A machine error data detection method based on the machine learning method according to Item 1. The step (d) includes
comparing a pattern of the error data with a pattern of the event data when error data exceeding the threshold data is detected at any one point in the operational data collected in real time;
and providing information on the presence or absence of a machine error and the presence or absence of an object defect in event data having a pattern corresponding to the error data to the operator terminal as information on the error data according to the comparison result. A machine error data detection method based on the described machine learning technique. The user interface provided to the worker terminal includes:
identification information about a plurality of machines included in the worksite;
including state information of the machine and the object manufactured by the machine that are matched and displayed for each identification information about each machine,
2. The method of detecting error data for machines based on machine learning techniques according to claim 1, wherein information about machine and object error data is provided as said state information when said error event occurs. The state information includes:
6. Machine error data based on the machine learning approach of claim 5 , comprising a state for machine normality and object normality, a state for machine normality and object malfunction, a state for machine error and object normality, and a state for machine error and object error. Detection method. 7. The machine of claim 6 , wherein real-time collected information regarding a plurality of chronological operational data for said machine is provided upon operator entry of identifying information for any one machine at said user interface. A machine error data detection method based on learning techniques. The chronological threshold data are
2. The machine error data detection method based on machine learning techniques of claim 1, wherein the motion data is collected in real time and updated by accumulating. The user interface provided to the worker terminal includes:
2. The machine error data detection method based on the machine learning technique of claim 1, comprising a graph of time-series operational data collected in real time, and the threshold data superimposed on the graph. In a server that detects machine error data based on machine learning techniques,
a memory storing a program for detecting machine error data based on a machine learning technique;
a processor for executing the program;
By executing the program, the processor
Collecting operational data over time for at least one machine;
dividing the motion data into predetermined time intervals, and mapping the divided motion data so as to overlap on the same time domain;
generating time-series threshold data by deriving time-series standard data for the set of mapped motion data based on machine learning;
If the operation data collected in real time exceeds the threshold data, it is determined as an error event, and information about the error event is provided to the worker terminal;
The step of collecting operational data over time for the at least one machine comprises:
Detecting event data having a pattern deviating from the average pattern of the collected motion data, matching and storing information regarding the presence or absence of errors in the machine and the presence or absence of defects in the object manufactured by the machine with respect to the event data. including the process
The mapping step includes:
A process of dividing the collected motion data by designating one cycle of the motion data as the predetermined time interval, and mapping the divided motion data on the time domain having a length corresponding to the one cycle. including
The process of providing information about the error event to the worker terminal includes:
If error data exceeding the threshold data is detected at any one point in the operation data collected in real time, it is determined whether there is an error in the machine and whether there is a defect in the object based on event data corresponding to the pattern of the error data. including the process
A server for detecting machine error data based on machine learning techniques, wherein one cycle of the motion data is the time it takes for the machine to manufacture one object. A computer-readable recording medium storing a computer program for causing the server to execute the machine error data detection method based on the machine learning technique according to claim 1. A computer program for causing the server to execute the machine error data detection method based on the machine learning technique according to claim 1.

Images