System and method for generating boundary of point cloud

Abstract

The present invention relates to a system and a method for generating a boundary of a point cloud, and provides a boundary generation system and an operation method thereof, the system comprising: a normal data reception unit for receiving multiple pieces of normal state data having multiple characteristic values; a cluster generation unit for generating multiple clusters by dividing the received pieces of normal state data; a boundary line configuration unit for configuring a boundary line of each cluster on the basis of the multiple characteristic values of the pieces of normal state data constituting each cluster; an abnormal data generation unit for generating virtual abnormal state data having multiple characteristic values outside the configured boundary line for each cluster; and a model training unit for generating an abnormality detection model by using the pieces of normal state data and the virtual abnormal state data as training data.

Description

Point cloud boundary generation system and method

[0001] The present invention relates to a point cloud boundary generation system and method, and more particularly, to a system and its operating method that can generate an artificial intelligence model capable of detecting whether a process is abnormal only with normal state data even when there is no abnormal state data in the process.

[0002] As artificial intelligence technology advances, smart factory technology is being activated to monitor various information on processes or equipment using sensors and detect or predict abnormal conditions based on artificial intelligence, thereby increasing process efficiency and minimizing the effort required for management.

[0003] However, to achieve this, it's necessary to distinguish between normal and abnormal data for each process or equipment, thereby providing training data. This process of categorizing data by status is called labeling. However, in many cases, equipment doesn't typically experience errors during initial operation. To address situations where errors occur due to aging, such situations must occur. Therefore, securing abnormal data for training, rather than normal data, presents challenges.

[0004] The prior art, Korean Patent No. 10-2433598, "Data Boundary Derivation System and Method," is a technology that generates learning data by distinguishing abnormal data from normal data in a state where normal data and abnormal data are not distinguished, and thereby creates an artificial intelligence model.

[0005] However, even when using such a technology, training data that distinguishes between normal and abnormal states can be generated only when abnormal data exists and is not labeled, so a technology is needed that can distinguish between normal and abnormal states, such as when normal data and abnormal data exist even when only normal data exists during normal operation.

[0006] The purpose of the present invention is to create an artificial intelligence model that can detect whether something is normal or not by virtually creating abnormal state data using only normal state data.

[0007] The purpose of the present invention is to create an artificial intelligence model that can accurately generate abnormal state data that is distinct from normal state data and detect whether something is normal or not even without abnormal state data.

[0008] The present invention aims to create an artificial intelligence model that can accurately detect whether something is normal or not without actual abnormal state data by creating abnormal state data that is not normal but does not deviate too much from the normal state.

[0009] The present invention aims to create an artificial intelligence model that can accurately detect whether something is normal or not even without actual abnormal state data by creating abnormal state data without bias.

[0010] In order to achieve this purpose, a boundary generation system according to an embodiment of the present invention may be configured to include a normal data receiving unit that receives a plurality of normal state data having a plurality of characteristic values, a cluster generating unit that generates a plurality of clusters by distinguishing the received normal state data, a boundary setting unit that sets a boundary of each cluster based on the plurality of characteristic values ​​of the normal state data constituting each cluster, an abnormal data generating unit that generates virtual abnormal state data having a plurality of characteristic values ​​outside the set boundary for each cluster, and a model learning unit that generates an abnormality detection model using the normal state data and the virtual abnormal state data as learning data.

[0011] At this time, the cluster generation unit may be configured to generate the plurality of clusters using an artificial intelligence model that learns the plurality of characteristic values ​​as input parameters to generate clusters, and the boundary line setting unit may be configured to set the boundary line based on the result value of the artificial intelligence model.

[0012] In addition, the above-described abnormal data generation unit can randomly generate virtual data and use the result value inputted into the artificial intelligence model that generates the cluster to select virtual data outside the set boundary as the virtual abnormal state data.

[0013] In addition, the above-described abnormal data generation unit may be configured to set a range using the minimum and maximum values ​​of characteristic values ​​of normal state data constituting each cluster with each of the plurality of characteristic values ​​as an axis, and to generate the virtual data within the range.

[0014] The present invention can achieve the effect of creating an artificial intelligence model capable of detecting whether something is normal or not by virtually creating abnormal state data using only normal state data.

[0015] The present invention can obtain the effect of detecting whether something is normal or not without the data of the abnormal state by accurately generating data of an abnormal state that is distinct from data of a normal state.

[0016] The present invention can achieve the effect of creating an artificial intelligence model that can accurately detect whether something is normal or not even without actual abnormal state data by creating abnormal state data that is not normal but does not deviate too much from the normal state.

[0017] The present invention can achieve the effect of generating an artificial intelligence model that can accurately detect whether something is normal or not even without actual abnormal state data by generating abnormal state data without bias.

[0018] FIG. 1 is a diagram illustrating the internal configuration of a boundary generation system according to one embodiment of the present invention.

[0019] FIG. 2 is a diagram illustrating an example of generating clusters by analyzing normal state data in a boundary generation system according to one embodiment of the present invention.

[0020] FIG. 3 is a drawing illustrating a range for generating virtual data in a boundary generation system according to an embodiment of the present invention.

[0021] FIG. 4 is a drawing illustrating an example of generating virtual abnormal state data around a boundary line in a boundary generation system according to one embodiment of the present invention.

[0022] FIG. 5 is a diagram illustrating an example of learning data generated in a boundary generation system according to an embodiment of the present invention.

[0023] FIG. 6 is a diagram illustrating an example of an anomaly detection model generated in a boundary generation system according to an embodiment of the present invention.

[0024] Figure 7 is a flowchart showing the flow of a boundary creation method according to one embodiment of the present invention.

[0025] In order to achieve this purpose, a boundary generation system according to an embodiment of the present invention may be configured to include a normal data receiving unit that receives a plurality of normal state data having a plurality of characteristic values, a cluster generating unit that generates a plurality of clusters by distinguishing the received normal state data, a boundary setting unit that sets a boundary of each cluster based on the plurality of characteristic values ​​of the normal state data constituting each cluster, an abnormal data generating unit that generates virtual abnormal state data having a plurality of characteristic values ​​outside the set boundary for each cluster, and a model learning unit that generates an abnormality detection model using the normal state data and the virtual abnormal state data as learning data.

[0026] At this time, the cluster generation unit may be configured to generate the plurality of clusters using an artificial intelligence model that learns the plurality of characteristic values ​​as input parameters to generate clusters, and the boundary line setting unit may be configured to set the boundary line based on the result value of the artificial intelligence model.

[0027] In addition, the above-described abnormal data generation unit can randomly generate virtual data and use the result value inputted into the artificial intelligence model that generates the cluster to select virtual data outside the set boundary as the virtual abnormal state data.

[0028] In addition, the above-described abnormal data generation unit may be configured to set a range using the minimum and maximum values ​​of characteristic values ​​of normal state data constituting each cluster with each of the plurality of characteristic values ​​as an axis, and to generate the virtual data within the range.

[0029] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. In describing the present invention, detailed descriptions of known components or functions will be omitted if they are deemed to obscure the gist of the invention. Furthermore, specific numerical values ​​used in describing embodiments of the present invention are merely exemplary and do not limit the scope of the invention.

[0030] The boundary generation system according to the present invention may be configured as a server equipped with a central processing unit (CPU) and memory, and capable of connecting to other terminals via a communication network such as the Internet. However, the present invention is not limited to the configuration of the central processing unit and memory. Furthermore, the boundary generation system according to the present invention may be physically configured as a single device, or may be implemented in a distributed manner across multiple devices.

[0031] In the present invention, both normal state data and abnormal state data are data detected from a process or equipment, etc., and are used as data to check and predict whether there is an error in the process or equipment from which the data was detected.

[0032] In the present invention, the normal state data and abnormal state data have multiple characteristic values, each of which represents a measurement item to be analyzed. When measured in a process or equipment, it may have values ​​such as temperature, voltage, and current. Using these characteristic values, an artificial intelligence model can be used to analyze whether a combination of characteristic values ​​at a specific point in time represents a normal or abnormal state of the process or equipment.

[0033] AI models are created by distinguishing between normal and abnormal data, inputting them as training data, and then training them. By inputting data representing the current state of a process or equipment into such an AI model, it can determine whether the current process or equipment is in a normal or abnormal state.

[0034] To achieve this, data for actual processes or equipment in normal and abnormal states must be prepared separately. However, if there were no abnormal situations, such as errors occurring in the process or equipment after the process or equipment was installed, data for abnormal states would not exist, making it impossible to create an artificial intelligence model that determines the state of the process or equipment.

[0035] In the present invention, even when only data derived from a normal state exists, virtual abnormal state data is generated based on this, thereby enabling the status of a process or equipment to be determined.

[0036] In addition to analyzing the status of processes or equipment, as in the examples above, the present invention can be applied to various fields, such as analyzing medical data to determine a patient's condition or identifying abnormal transactions using financial data. The definitions of the characteristic values ​​constituting each piece of data may vary depending on the application. The present invention is not limited by the types of characteristic values ​​of such data.

[0037] FIG. 1 is a drawing showing the configuration of an inspection device according to one embodiment of the present invention.

[0038] As illustrated in the drawing, a boundary generation system (101) according to one embodiment of the present invention may be configured to include a normal data receiving unit (110), a cluster generating unit (120), a boundary line setting unit (130), an abnormal data generating unit (140), and a model learning unit (150).

[0039] Each component may be a software module that operates within the same physical device, and may be configured so that two or more physically separated devices can operate in conjunction with each other through a communication network or the like. Various embodiments that include the same function fall within the scope of the present invention.

[0040] The normal data receiving unit (110) receives a plurality of normal state data having a plurality of characteristic values. The normal state data is data that can confirm the status of a process or equipment that is the target for monitoring an abnormal state. As described above, various measurement data such as temperature, humidity, voltage, and current that can confirm the status can be included in the plurality of characteristic values.

[0041] The normal data receiving unit (110) may be data generated by storing characteristic values ​​measured by installing sensors on the process or equipment being monitored in a time-series manner and grouping them into time units. The normal data received in this way is data with dimensions corresponding to the number of characteristic values. That is, if there are two characteristic values, it is two-dimensional data, and if there are five characteristic values, it is five-dimensional data.

[0042] The normal state data received by the normal data receiver (110) is data collected during the normal operation of the process or equipment being monitored. When monitoring a monitoring target, especially a newly installed process or equipment, there are often no errors that have yet occurred. Therefore, the present invention collects only data during normal operation, allowing for the determination of whether the monitoring target is in a normal or abnormal state.

[0043] The cluster generation unit (120) classifies the received normal state data and creates multiple clusters. When analyzing the characteristic values ​​of the normal state data, clusters can be formed by grouping data belonging to categories with similar characteristic values. One cluster or multiple clusters may be formed.

[0044] The cluster generation unit (120) generates the plurality of clusters using an artificial intelligence model that learns the plurality of characteristic values ​​as input parameters and generates clusters. Various methods can be used to generate clusters of data with similar characteristics using multidimensional data as input data, and artificial intelligence technologies such as machine learning or deep learning can be applied for this purpose.

[0045] When a cluster is created in the cluster creation unit (120), a probability value for each data item to be included in each cluster is derived. By using this probability value, each data item can be distinguished to which cluster it belongs, thereby forming a cluster.

[0046] In the process of creating a cluster in the cluster creation unit (120), there are cases where anomaly detection (PCA, Auto Encoder) technology based on restoration error using dimension reduction is applied, but in this case, information loss may occur, making accurate anomaly detection difficult. Therefore, it is desirable to form a cluster in a way that maintains the input data as is.

[0047] For this purpose, the cluster generation unit (120) can utilize various clustering techniques such as Gaussian Mixture Model, Agglomerative Clustering, Spectral Clustering, and Umap Clustering.

[0048] The boundary setting unit (130) sets the boundary of each cluster based on the plurality of characteristic values ​​of the normal state data constituting each cluster. The boundary may be referred to as a boundary surface depending on the number of dimensions of the data, and serves as a criterion for determining the range of normal state data. Therefore, when the boundary is set, data within the boundary can be classified as normal state data, and data outside the boundary can be classified as abnormal state data. Since only normal state data is input, abnormal state data is generated outside the boundary to enable the normal state and abnormal state to be distinguished.

[0049] The boundary setting unit (130) sets the boundary based on the result values ​​of the artificial intelligence model. The boundary of each cluster can be determined based on the probability value of belonging to the corresponding cluster provided for each data in the artificial intelligence model that creates the cluster as described above.

[0050] The abnormal data generation unit (140) generates virtual abnormal state data having multiple characteristic values ​​outside the set boundary line for each cluster. As described above, since the boundary line is set to confirm the range of normal state data, if any data is generated outside the boundary line, it can be regarded as abnormal state data. Therefore, virtual abnormal state data can be generated outside the boundary line for the multiple generated clusters. The abnormal state data has the same type of characteristic values ​​as the normal state data, and by virtually setting these characteristic values, virtual abnormal state data can be generated.

[0051] The abnormal data generation unit (140) randomly generates virtual data, and uses the result values ​​inputted into the artificial intelligence model that generates the clusters to select virtual data outside the set boundary line as the virtual abnormal state data. Rather than simply randomly generating virtual abnormal state data outside the boundary line, by using the probability value, which is the result value derived from the artificial intelligence model that determines the clusters, the virtual abnormal state data can be generated so as to be evenly distributed around the boundary line.

[0052] In order to ensure that the abnormal data generation unit (140) can be evenly distributed around the boundary line, it is preferable to input the randomly generated virtual data into an artificial intelligence model that generates clusters, and select data whose probability value, which is the result value derived from being input into the artificial intelligence model, is within a predetermined range as virtual abnormal state data among the input virtual data. If abnormal state data having characteristics that are too different from each cluster or having characteristics of only a specific region is virtually generated, it is difficult to accurately distinguish the actual abnormal state. Therefore, it is preferable to generate the virtual abnormal state data so that it is evenly distributed without going far from the generated boundary line.

[0053] The abnormal data generation unit (140) sets a range using the minimum and maximum values ​​of the characteristic values ​​of the normal state data constituting each cluster based on each of the plurality of characteristic values ​​as an axis, and generates the virtual data within the range. When generating virtual data outside the boundary line, since the outside of the boundary line is virtually infinite, if virtual data is generated randomly, meaningless virtual data is generated and unnecessary operations are required to exclude it. Therefore, it is preferable to set the range of virtual data using the characteristic values ​​of the normal state data constituting the cluster, so that virtual data is generated near the boundary line, and appropriate data among them can be generated as virtual abnormal state data.

[0054] The model learning unit (150) creates an anomaly detection model using the normal state data and the virtual abnormal state data as learning data. As described above, once the normal state data and the virtual abnormal state data are secured, an anomaly detection model can be created using an artificial intelligence algorithm using these as learning data. At this time, various machine learning or deep learning-based artificial intelligence models used when actual abnormal state data is secured can be applied, and the present invention is not limited by such methods.

[0055] The artificial intelligence model generated through learning in the model learning unit (150) can be used to monitor the status of the process or equipment to be monitored thereafter and detect or predict abnormal conditions when they occur.

[0056] FIG. 2 is a diagram illustrating an example of generating clusters by analyzing normal state data in a boundary generation system according to one embodiment of the present invention.

[0057] In order to improve understanding, the drawing assumes that the normal state data is two-dimensional data and expresses it. As described above, in the case of two-dimensional data, each data has two characteristic values.

[0058] The circled dots in the diagram represent normal-state data. When clustering normal-state data, multiple clusters are created, as shown in the diagram. The diagram shows three clusters, all of which are normal, but data with different characteristics can be derived depending on the situation. Therefore, clusters can be created based on these characteristics.

[0059] The lines drawn on the boundaries of the three clusters in the drawing are the boundaries that separate the clusters. The boundaries can be derived using the probability values, which are the output values ​​derived from the AI ​​model for cluster creation. Through this, the area inside the boundary line can be distinguished as normal state data, and the area outside the boundary line as abnormal state data. By virtually generating abnormal state data outside the boundary line, it is possible to create an AI model that can detect whether the monitoring target is abnormal even when there is no abnormal state data indicating that the actual monitoring target is behaving abnormally.

[0060] FIG. 3 is a drawing illustrating a range for generating virtual data in a boundary generation system according to an embodiment of the present invention.

[0061] As illustrated in the diagram, once a cluster is created, the range is set using the characteristic values ​​of the data constituting each cluster. When the range is set using the minimum and maximum values ​​of each characteristic value, a range is set, as indicated by the rectangle in the diagram.

[0062] Since the drawing uses two-dimensional data as an example, a rectangular range is set. As the number of characteristic values ​​increases and the dimensionality increases, a range appropriate to that dimension is set. Setting the range in this way allows for the virtual generation of abnormal state data that falls within the range but outside the boundary, thereby building learning data capable of distinguishing between actual normal and virtual states.

[0063] FIG. 4 is a drawing illustrating an example of generating virtual abnormal state data around a boundary line in a boundary generation system according to one embodiment of the present invention.

[0064] The drawing is an example of one cluster in the upper left corner among the three clusters in Fig. 3. By creating virtual abnormal state data around the boundary line among the areas outside the boundary line and within the set range as indicated by the star shape in the drawing, learning data that can detect abnormal states can be created.

[0065] In order to generate abnormal state data around the boundary line, virtual data can be input into an artificial intelligence model that generates clusters as described above, and configured to select only virtual data whose probability value falls within a predetermined range outside the boundary line as virtual abnormal state data.

[0066] FIG. 5 is a diagram illustrating an example of learning data generated in a boundary generation system according to an embodiment of the present invention.

[0067] As shown in the drawing, when virtual abnormal state data is generated, the actual normal state data and the virtual abnormal state data are combined, and the actual normal state data is labeled as normal state and the virtual abnormal state data is labeled as abnormal state to generate learning data.

[0068] Through this process, we can create an anomaly detection model that can distinguish between normal and abnormal conditions without actually having abnormal condition data.

[0069] FIG. 6 is a diagram illustrating an example of an anomaly detection model generated in a boundary generation system according to an embodiment of the present invention.

[0070] By training an artificial intelligence algorithm using the learning data generated in Fig. 5, an anomaly detection model that can detect anomalies by distinguishing between normal and abnormal states of the monitoring target can be created.

[0071] The generated anomaly detection model distinguishes between normal and abnormal areas, as indicated by the circular lines in the drawing, thereby distinguishing whether the actual data collected from the monitoring target is normal or abnormal data, and thereby determining whether the monitoring target is operating normally.

[0072] Figure 7 is a flowchart showing the flow of an inspection method according to one embodiment of the present invention.

[0073] A boundary generation method according to one embodiment of the present invention relates to an operating method of a boundary generation system (101) having a central processing unit and a memory, and the description of the boundary generation system (101) described above can be applied as is.

[0074] Therefore, even if there is no separate explanation below, it is self-evident that all contents described to explain the boundary generation system (101) can be applied as is to implementing the inspection method.

[0075] The normal data reception step (S701) receives a plurality of normal state data having multiple characteristic values. The normal state data is data that can confirm the status of a process or equipment that is the target for monitoring an abnormal state. As previously described, various measurement data such as temperature, humidity, voltage, and current that can confirm the status can be included in the plurality of characteristic values.

[0076] The normal data reception step (S701) may be data generated by installing sensors on the process or equipment being monitored, storing the measured characteristic values ​​in a time-series fashion, and grouping them by time unit. The normal data received in this manner has dimensions corresponding to the number of characteristic values. That is, if there are two characteristic values, the data is two-dimensional, and if there are five characteristic values, the data is five-dimensional.

[0077] The normal state data received in the normal data reception step (S701) is data collected during the normal operation of the process or equipment being monitored. When monitoring a new process or equipment, it is often the case that no error conditions have yet occurred. Therefore, the present invention collects only data from normally operating situations, allowing for the determination of whether the monitoring target is in a normal or abnormal state.

[0078] The cluster creation step (S702) classifies the received normal state data and creates multiple clusters. When analyzing the characteristic values ​​of normal state data, clusters can be formed by grouping data belonging to categories with similar characteristic values. A single cluster or multiple clusters may be formed.

[0079] The cluster creation step (S702) generates the multiple clusters using an artificial intelligence model that learns the multiple characteristic values ​​as input parameters and creates clusters. Various methods can be used to create clusters of data with similar characteristics using multidimensional data as input data, and artificial intelligence technologies such as machine learning or deep learning can be applied for this purpose.

[0080] When creating clusters in the cluster creation step (S702), a probability value for each data item is derived to determine its inclusion in each cluster. Using these probability values, clusters can be formed by identifying which cluster each data item belongs to.

[0081] In the process of creating clusters in the cluster creation step (S702), there are cases where anomaly detection based on restoration error using dimensionality reduction (PCA, Auto Encoder) technology is applied. However, in this case, information loss may occur, making accurate anomaly detection difficult. Therefore, it is desirable to form clusters in a way that maintains the input data as is.

[0082] For this purpose, the cluster creation step (S702) can utilize various clustering techniques such as Gaussian Mixture Model, Agglomerative Clustering, Spectral Clustering, and Umap Clustering.

[0083] The boundary setting step (S703) sets the boundary of each cluster based on the multiple characteristic values ​​of the normal state data constituting each cluster. The boundary can also be called a boundary surface depending on the number of data dimensions, and serves as a criterion for determining the range of normal state data. Therefore, once the boundary is set, data within the boundary can be classified as normal state data, and data outside the boundary can be classified as abnormal state data. Since only normal state data is input, abnormal state data is generated outside the boundary to enable the distinction between normal and abnormal states.

[0084] The boundary setting step (S703) sets the boundary based on the output values ​​of the artificial intelligence model. The boundary of each cluster can be determined based on the probability value provided for each data item belonging to the cluster by the artificial intelligence model that creates the clusters, as described above.

[0085] The abnormal data generation step (S704) generates virtual abnormal state data having multiple characteristic values ​​outside the set boundary line for each cluster. As described above, since the boundary line is set to confirm the range of normal state data, if any data is generated outside the boundary line, it can be viewed as abnormal state data. Therefore, virtual abnormal state data can be generated outside the boundary line for the multiple generated clusters. The abnormal state data has the same type of characteristic values ​​as the normal state data, and by setting these characteristic values ​​virtually, virtual abnormal state data can be generated.

[0086] The abnormal data generation step (S704) randomly generates virtual data, and uses the result values ​​inputted into the artificial intelligence model that generates the clusters to select virtual data outside the set boundary as the virtual abnormal state data. Rather than simply randomly generating virtual abnormal state data outside the boundary, by using the probability value, which is the result value derived from the artificial intelligence model that determines the clusters, the virtual abnormal state data can be generated so that it is evenly distributed around the boundary.

[0087] In order to ensure that the abnormal data is evenly distributed near the boundary line, the abnormal data generation step (S704) preferably inputs the randomly generated virtual data into an artificial intelligence model that generates clusters, and selects data whose probability values, which are the result values ​​derived from the input virtual data and inputted into the artificial intelligence model, are within a predetermined range as the virtual abnormal state data. If abnormal state data having characteristics that are too different from each cluster or having characteristics of only a specific region are virtually generated, it is difficult to accurately distinguish the actual abnormal state. Therefore, it is preferable to generate the virtual abnormal state data so that it is evenly distributed without going far from the generated boundary line.

[0088] The abnormal data generation step (S704) sets a range using the minimum and maximum values ​​of the characteristic values ​​of the normal state data constituting each cluster based on each of the plurality of characteristic values ​​as an axis, and generates the virtual data within the range. When generating virtual data outside the boundary line, since the outside of the boundary line is virtually infinite, if virtual data is generated randomly, meaningless virtual data is generated and unnecessary operations are required to exclude it. Therefore, it is preferable to set the range of virtual data using the characteristic values ​​of the normal state data constituting the cluster, so that virtual data is generated near the boundary line, and appropriate data among them can be generated as virtual abnormal state data.

[0089] The model learning step (S705) creates an anomaly detection model using the normal state data and the virtual abnormal state data as learning data. As described above, once the normal state data and the virtual abnormal state data are secured, an anomaly detection model can be created using an artificial intelligence algorithm using these as learning data. At this time, various machine learning or deep learning-based artificial intelligence models used when actual abnormal state data is secured can be applied, and the present invention is not limited by such methods.

[0090] The artificial intelligence model generated through learning in the model learning step (S705) can be used to monitor the status of the process or equipment to be monitored thereafter and detect or predict abnormal conditions when they occur.

[0091] The boundary generation method according to the present invention can be produced as a program for causing a computer to execute the program and recorded on a computer-readable recording medium.

[0092] Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical recording media such as CDROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, and flash memory.

[0093] Examples of program instructions include not only machine language codes, such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform processing according to the present invention, and vice versa.

[0094] Although the present invention has been described above with reference to embodiments, those skilled in the art can modify and change the present invention in various ways without departing from the spirit and scope of the present invention as set forth in the following claims.

[0095] The present invention relates to a point cloud boundary generation system and method, and provides a boundary generation system and an operating method thereof, including a normal data receiving unit that receives a plurality of normal state data having a plurality of characteristic values, a cluster generating unit that generates a plurality of clusters by distinguishing the received normal state data, a boundary setting unit that sets a boundary of each cluster based on the plurality of characteristic values ​​of the normal state data constituting each cluster, an abnormal data generating unit that generates virtual abnormal state data having a plurality of characteristic values ​​outside the set boundary for each cluster, and a model learning unit that generates an abnormality detection model using the normal state data and the virtual abnormal state data as learning data.

Claims

1. A normal data receiving unit that receives a plurality of normal state data having a plurality of characteristic values; A cluster generation unit that generates multiple clusters by distinguishing the received normal state data; A boundary setting unit that sets a boundary of each cluster based on the plurality of characteristic values ​​of the normal state data constituting each cluster; An abnormal data generation unit that generates virtual abnormal state data having multiple characteristic values ​​outside the set boundary line for each of the above clusters, and A model learning unit that creates an anomaly detection model using the above normal state data and the above virtual abnormal state data as learning data. A boundary generation system including:

2. In paragraph 1, The above cluster generation unit Generating the plurality of clusters using an artificial intelligence model that learns the plurality of characteristic values ​​as input parameters and generates clusters, The above boundary setting section Setting a boundary line based on the result value of the above artificial intelligence model A boundary generation system featuring:

3. In paragraph 2, The above abnormal data generation unit Randomly generate virtual data, Selecting virtual data outside the set boundary as the virtual abnormal state data by using the result value input into the artificial intelligence model that creates the cluster of the generated virtual data. A boundary generation system featuring:

4. In paragraph 3, The above abnormal data generation unit Setting a range using the minimum and maximum values ​​of the characteristic values ​​of the normal state data constituting each cluster with each of the above multiple characteristic values ​​as an axis, and generating the virtual data within the range. A boundary generation system featuring:

5. A boundary generation method operating in a boundary generation system having a central processing unit and memory, A normal data receiving step for receiving a plurality of normal state data having a plurality of characteristic values; A cluster creation step for creating multiple clusters by distinguishing the normal state data received above; A boundary setting step for setting a boundary of each cluster based on the plurality of characteristic values ​​of the normal state data constituting each cluster; An abnormal data generation step for generating virtual abnormal state data having multiple characteristic values ​​outside the set boundary line for each of the above clusters, and A model learning step for creating an anomaly detection model using the above normal state data and the above virtual abnormal state data as learning data. A method for generating boundaries including .

6. In paragraph 5, The above cluster creation steps are Generating the plurality of clusters using an artificial intelligence model that learns the plurality of characteristic values ​​as input parameters and generates clusters, The above boundary setting step is Setting a boundary line based on the result value of the above artificial intelligence model A method for generating boundaries characterized by .

7. In paragraph 6, The above abnormal data generation step is Randomly generate virtual data, Selecting virtual data outside the set boundary as the virtual abnormal state data by using the result value input into the artificial intelligence model that creates the cluster of the generated virtual data. A method for generating boundaries characterized by .

8. In paragraph 7, The above abnormal data generation step is Setting a range using the minimum and maximum values ​​of the characteristic values ​​of the normal state data constituting each cluster with each of the above multiple characteristic values ​​as an axis, and generating the virtual data within the range. A method for generating boundaries characterized by .

9. A computer-readable recording medium having recorded thereon a program for causing a computer to execute the method of any one of clauses 5 to 8.

Images