Method and apparatus for detecting anomaly of electric rod in electric furnace

Abstract

Provided are a device and a method for detecting an anomaly of an electrode rod through the steps of: collecting time series data from a plurality of sensors used for process monitoring of an electric furnace; reconstructing input time series data of a predetermined size by using a reconstruction-based artificial neural network model; calculating a reconstruction error between the input time series data and the reconstructed time series data; and determining, on the basis of the reconstruction error, whether the electrode rod is damaged.

Description

Method and device for detecting abnormalities in electrode rods of electric furnaces

[0001] This invention relates to a method and device for detecting an abnormality in an electrode rod of an electric furnace.

[0002] Electric furnaces (EFFs) use heat generated by an electric arc to melt metal or ore, producing molten iron. The electric arc can occur between an electrode and metal (such as scrap) within the furnace. The electrode efficiently generates arc heat through optimal positioning control. Typically, the position of the electrode in an EF can be controlled through impedance and constant current control.

[0003] Electrode damage can occur for several reasons during the electric furnace process. Primary causes include collisions between the electrode and a non-conductor, resonance of the electrode due to arc current, collisions between the electrode and the molten metal, and vibration of the electrode support (electrode clamp). Because the internal temperature of the electric furnace exceeds 1,200°C, camera installation is difficult, making visual observation of electrode damage difficult.

[0004] If the process continues without addressing electrode damage, a major accident can occur, and repair costs increase over time. For example, if just four seconds pass after an electrode break, a high-current arc can travel to equipment surrounding the electrode, potentially causing a fire.

[0005] One embodiment provides a method for detecting an abnormality in an electrode rod of an electric furnace.

[0006] Another embodiment provides a method for detecting an abnormality in an electrode rod of an electric furnace.

[0007] Another embodiment provides a device for detecting an abnormality in an electrode rod of an electric furnace.

[0008] According to one embodiment, a method for detecting an abnormality in an electrode of an electric furnace is provided. The method comprises the steps of collecting time series data indicating the condition of the electrode, restoring input time series data within the collected time series dataset using an autoencoder model, calculating a restoration error between the input time series data and the restored time series data, and determining whether the electrode is damaged based on a comparison result between the restoration error and a reference value.

[0009] In the above method, the step of collecting time series data indicating the state of the electrode rod may include the step of collecting time series data from a plurality of sensors used for process monitoring of the electric furnace, and the step of collecting position control values ​​used for position control of the electrode rod.

[0010] In the above method, the step of calculating a restoration error between the input time series data and the restored time series data may include a step of calculating a restoration error for each of a plurality of electrode rods included in the electric furnace.

[0011] In the above method, the autoencoder model may be pre-trained to restore time series data similar to normal time series data by using normal time series data collected when the electrode rod is not damaged as training data.

[0012] In the above method, the reference value can be determined in advance by considering the process progress and the average lifespan of the electrode rod.

[0013] In the above method, the step of determining whether the electrode rod is damaged based on the comparison result between the restoration error and the reference value includes the step of determining that the electrode rod is damaged when n or more input time series data having a restoration error greater than the reference value are continuously detected within the time series data set, where n may be a natural number smaller than the size of the time series data set.

[0014] In the above method, the step of determining whether the electrode rod is damaged based on the comparison result between the restoration error and the reference value includes the step of determining that the electrode rod is damaged when m or more input time series data having a restoration error greater than the reference value are detected within the time series data set, where m may be a natural number smaller than the size of the time series data set.

[0015] In the above method, if input time series data having a restoration error greater than an auxiliary reference value is detected within the time series dataset, a step of preliminarily determining that damage will soon occur to the electrode rod is further included, wherein the auxiliary reference value may be smaller than the reference value.

[0016] According to another embodiment, a method for detecting an abnormality in an electrode of an electric furnace is provided. The method includes the steps of collecting time series data indicating the state of the electrode, restoring input time series data within the collected time series dataset, calculating a restoration error between the input time series data and the restored time series data, calculating a correlation inconsistency of the input time series data, calculating an abnormality score of the input time series data based on the restoration error and the correlation inconsistency, and determining whether the electrode is damaged based on a comparison result between the abnormality score and a reference score.

[0017] In the above method, the step of restoring the input time series data and calculating the restoration error between the input time series data and the restored time series data may include the step of dividing the collected time series data into a predetermined size and inputting the data into an anomaly transformer model.

[0018] In the above method, the step of calculating the association inconsistency of the input time series data may include the step of calculating the association inconsistency using the prior association and series association calculated in the anomaly attention layer of the anomaly transformer.

[0019] In the above method, the anomaly transformer model may be pre-trained to restore time series data similar to normal time series data by using normal time series data collected when the electrode rod is not damaged as training data.

[0020] In the above method, the step of calculating an anomaly score of input time series data based on the restoration error and the association mismatch may include a step of calculating a product of the restoration error and the association mismatch as the anomaly score.

[0021] In the above method, the reference score can be determined in advance by considering the process progress and the average life of the electrode rod.

[0022] In the above method, the step of determining whether the electrode rod is damaged based on the comparison result between the abnormal score and the reference score includes the step of determining that the electrode rod is damaged when n or more input time series data having an abnormal score greater than the reference score are continuously detected within the time series data set, where n may be a natural number smaller than the size of the time series data set.

[0023] In the above method, the step of determining whether the electrode rod is damaged based on the comparison result between the abnormal score and the reference score includes the step of determining that the electrode rod is damaged when m or more input time series data having an abnormal score greater than the reference score are detected within the time series data set, where m may be a natural number smaller than the size of the time series data set.

[0024] The method further includes a step of preliminarily determining that damage will soon occur to the electrode rod if input time series data having an abnormality score greater than an auxiliary reference score is detected within the time series dataset, wherein the auxiliary reference score may be less than the reference score.

[0025] According to another embodiment, a device for detecting an abnormality in an electrode of an electric furnace is provided. The device includes a processor and a memory, and the memory stores instructions for causing the processor to execute a plurality of steps, the plurality of steps including: collecting time series data from a plurality of sensors used for process monitoring of the electric furnace; restoring input time series data having a predetermined size using an artificial neural network model based on a restoration method; calculating a restoration error between the input time series data and the restored time series data; and determining whether the electrode is damaged based on the restoration error.

[0026] In the above device, the artificial neural network model of the restoration-based method is an autoencoder model, and the step of determining whether the electrode rod is damaged based on the restoration error may include a step of determining that the electrode rod is damaged when the restoration error is greater than a reference value.

[0027] In the above device, the artificial neural network model of the restoration-based method is an anomaly transformer model, and the plurality of steps may further include a step of calculating an association inconsistency of input time series data, and a step of calculating an anomaly score of input time series data based on the restoration error and the association inconsistency.

[0028] In the above device, the step of determining whether the electrode rod is damaged based on the restoration error may include a step of determining whether the electrode rod is damaged based on the comparison result between the abnormal score and the reference score.

[0029] In the above device, the plurality of sensors may include a current sensor for measuring a current supplied to the electrode rod, a voltage sensor for measuring a voltage formed on the electrode rod, and an impedance sensor for measuring an impedance of the electrode rod.

[0030] By utilizing an artificial neural network model based on restoration, abnormalities in electrodes can be detected in real time, thereby preventing accidents caused by electrode damage and significantly reducing time and cost.

[0031] Figure 1 is a drawing showing an electric furnace and an electrode rod according to one embodiment.

[0032] FIG. 2 is a block diagram showing an abnormality detection device for detecting an abnormality of an electrode rod according to one embodiment.

[0033] FIG. 3 is a flowchart illustrating a method for detecting an abnormality in an electrode rod using an autoencoder model according to one embodiment.

[0034] Figures 4 and 5 are diagrams showing an autoencoder model according to one embodiment.

[0035] FIG. 6 is a flowchart illustrating a method for detecting anomalies in electrode rods using an anomaly transformer model according to another embodiment.

[0036] Fig. 7 is a diagram showing an anomaly transformer model according to another embodiment.

[0037] Fig. 8 is a block diagram showing an abnormality detection device of an electrode rod according to another embodiment.

[0038] Below, with reference to the attached drawings, embodiments of the present disclosure are described in detail so that those skilled in the art can easily practice the present disclosure. However, the present disclosure may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, irrelevant parts have been omitted for clarity, and similar reference numerals have been used throughout the specification to indicate similar parts.

[0039] In this description, each of the phrases "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" may include any one of the items listed together in that phrase, or all possible combinations thereof.

[0040] In this description, when a part is said to "include" a certain component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise specifically stated.

[0041] In this document, expressions written in the singular may be interpreted as singular or plural, unless explicit expressions such as “one” or “single” are used.

[0042] As used herein, “and / or” includes each and every combination of one or more of the mentioned components.

[0043] In this disclosure, terms including ordinal numbers, such as "first" and "second," may be used to describe various components, but these components are not limited by these terms. These terms are used solely to distinguish one component from another. For example, without departing from the scope of the present disclosure, a first component could be referred to as a "second component," and similarly, a second component could also be referred to as a "first component."

[0044] In the flowcharts described herein with reference to the drawings, the order of operations may be changed, several operations may be merged, some operations may be split, and certain operations may not be performed.

[0045] The artificial intelligence model (AI model) of the present disclosure is a machine learning model that learns at least one task and can be implemented as a computer program executed by a processor. The task learned by the AI ​​model may refer to a problem to be solved through machine learning or a task to be performed through machine learning. The AI ​​model can be implemented as a computer program executed on a computing device, downloaded over a network, or sold as a product. Alternatively, the AI ​​model can be connected to various devices over a network.

[0046] Figure 1 is a drawing showing an electric furnace and an electrode rod according to one embodiment.

[0047] In one embodiment, an electric furnace (1) is a furnace that melts metal or ore using heat generated by electricity. In the electric furnace process, an arc discharge is generated between the electrodes (2) and the molten material by controlling the positions of a plurality of electrode rods (2) to which a high voltage is applied, and the molten material can be melted by the heat of the arc discharge to form molten steel (3). Thereafter, the molten steel (3) can be sent out of the electric furnace (1) through the tapping port (4).

[0048] An abnormality detection device for an electrode rod (2) according to one embodiment can determine whether the electrode rod (2) is damaged by using an artificial neural network model based on a restoration-based method based on time series data acquired from a plurality of sensors for monitoring an electric furnace process. Below, an abnormality detection method of an abnormality detection device for an electrode rod (2) according to one embodiment is described in detail with reference to FIGS. 2 to 7.

[0049] FIG. 2 is a block diagram showing an abnormality detection device for detecting an abnormality of an electrode rod according to one embodiment.

[0050] Referring to FIG. 2, an anomaly detection device (100) according to one embodiment may include a data collector (110), a data preprocessor (120), and an anomaly determiner (130).

[0051] In one embodiment, the data collector (110) can collect time-series data from multiple sensors used for process monitoring of the electric furnace. The multiple sensors can include a current sensor for measuring the current supplied to the electrode (2), a voltage sensor for measuring the voltage formed on the electrode (2), and an impedance sensor for measuring the impedance of the electrode (2). The data collector (110) can also collect position control values ​​used for position control of the electrode (2).

[0052] In one embodiment, the data preprocessor (120) may divide time series data into predetermined sizes and input a time series dataset having the predetermined sizes to the anomaly detector (130). For example, when a sensor samples data at intervals of 20 ms, the data preprocessor (120) may generate a time series dataset of size 10, including 10 time series data collected over 0.2 seconds, as input to the anomaly detector (130).

[0053] In one embodiment, the abnormality determiner (130) can determine an abnormality of the electrode (2) from preprocessed time series data using an artificial neural network model in a reconstruction-based manner. In one embodiment, the abnormality determiner (130) can use an auto encoder (AE) model to detect an abnormality of the electrode (2). Alternatively, the abnormality determiner (130) can use a transformer model to detect an abnormality of the electrode (2). In one embodiment, the abnormality determiner (130) can detect an abnormality of the electrode (2) by processing time series data in an unsupervised learning manner.

[0054] FIG. 3 is a flowchart illustrating a method for detecting an abnormality of an electrode rod using an autoencoder model according to one embodiment, and FIGS. 4 and 5 are diagrams illustrating an autoencoder model according to one embodiment.

[0055] Referring to FIG. 3, time series data collected by a data collector (110) and preprocessed to a predetermined size by a data preprocessor (120) are input to an anomaly determination unit (130) (S110). The size of the preprocessed time series data may be predetermined, for example, to 256 (N=256).

[0056] Referring to FIG. 3, the anomaly determination unit (130) can generate restored time series data corresponding to the input time series data and calculate a restoration error between the input time series data and the restored time series data (S120). The anomaly determination unit (130) can calculate the restoration error of the time series data for each electrode rod (2) included in the electric furnace (1). For example, when the electric furnace (1) includes three electrode rods (2), the anomaly determination unit (130) can calculate a restoration error of a different value for each electrode rod (2).

[0057] Referring to FIG. 4, the anomaly detector (130) can generate restored time series data corresponding to the input time series data using an autoencoder model (131) including an encoder and a decoder. In one embodiment, the autoencoder model (131) may be pre-trained to restore time series data similar to the normal time series data by using normal time series data collected when the electrode rod (2) is not damaged as training data. Accordingly, when normal time series data having high similarity to each other are input, the autoencoder model (131) can restore time series data similar to the input, but when abnormal time series data having low similarity to the time series data used for pre-training is input, it cannot perform restoration similar to the input. In addition, when the autoencoder model (131) restores time series data that is not similar to the input, the restoration error between the input time series data and the restored time series data may increase.

[0058] In one embodiment, when a time series dataset of a predetermined size is input to an anomaly detector (130), the anomaly detector (130) can perform restoration of the time series dataset in parallel using an autoencoder model (131) and calculate the restoration error between the restored time series dataset and the input time series dataset at the same time. For example, the anomaly detector (130) can perform restoration in parallel for 10 time series data within a time series dataset of size 10 collected for 0.2 seconds and simultaneously determine 10 restoration errors between the restored time series data and the corresponding input time series data.

[0059] Referring to FIG. 5, the structure of the autoencoder model (131) used by the anomaly detector (130) is illustrated. The encoder of the autoencoder model (131) may include three convolutional neural network (CNN) layers, and the decoder of the autoencoder model (131) may include four CNN layers. The filter size is 7, and the Relu function may be used as the activation function applied to each CNN. In addition, the Adam optimizer may be used as an optimal hyperparameter for pre-learning.

[0060] Referring to FIG. 3, the abnormality determination unit (130) can determine whether the electrode rod (2) is damaged based on the comparison result between the restoration error of time series data having a predetermined size and a reference value (S130). The reference value for the restoration error can be determined in advance by considering the process progress, the average lifespan of the electrode rod (2), etc.

[0061] For example, if the restoration error of the time series data does not exceed 1% (reference value) or is less than 1%, the abnormality determiner (130) may determine that there is no damage to the electrode rod (2), and if the restoration error of the time series data exceeds 1% or is 1% or more, the abnormality determiner (130) may determine that there is damage to the electrode rod (2).

[0062] When a time series data set of a predetermined size is input to an anomaly judger (130), the anomaly judger (130) can judge the occurrence of damage to the electrode rod (2) based on whether there is time series data in the time series data set whose restoration error with respect to the restored time series data exceeds 1%.

[0063] In one embodiment, the abnormality determiner (130) can determine that damage has occurred to the electrode rod (2) when n or more consecutive time series data having a restoration error greater than a reference value exist within a time series data set of a predetermined size, or m or more time series data having a restoration error greater than a reference value exist. In this case, both m and n are natural numbers smaller than the size of the time series data set, and m may be greater than n.

[0064] For example, if six or more time series data having a restoration error of 1% or more are detected within a time series dataset of size 10, or three or more are detected consecutively, the abnormality determiner (130) can determine that damage has occurred to the electrode rod (2).

[0065] Alternatively, the abnormality judgement device (130) may preliminarily judge that damage will soon occur to the electrode rod (2) if time series data having a restoration error greater than an auxiliary reference value is detected within a time series data set of a predetermined size. In this case, the auxiliary reference value may be determined in advance to be smaller than the reference value (e.g., 80% of the reference value) by taking into account the process progress, the average lifespan of the electrode rod (2), etc.

[0066] In one embodiment, the abnormality determiner (130) can preliminarily determine that damage will soon occur to the electrode rod (2) when n or more consecutive time series data having a restoration error greater than an auxiliary reference value exist within a time series data set of a predetermined size, or m or more time series data having a restoration error greater than an auxiliary reference value exist. In this case, both m and n are natural numbers smaller than the size of the time series data set, and m may be greater than n.

[0067] For example, if six or more time series data having a restoration error of 0.8% or more are detected within a time series dataset of size 10, or three or more are detected consecutively, the abnormality determiner (130) can preliminarily determine that damage will soon occur to the electrode rod (2).

[0068] If the restoration error of the time series data exceeds a predetermined reference value, the abnormality determiner (130) may determine that the electrode rod (2) is damaged and may notify of damage to the electrode rod (2) (S140). Alternatively, if the restoration error of the time series data exceeds a predetermined auxiliary reference value, the abnormality determiner (130) may preliminarily determine that the electrode rod (2) is about to be damaged and may notify of a risk of damage to the electrode rod (2). The notification of damage to the electrode rod (2) and / or the notification of a risk of damage to the electrode rod (2) may be performed by various methods (visual display, voice notification, etc.) to the control subject of the electric furnace (1).

[0069] As described above, by utilizing the autoencoder model to detect abnormalities in the electrode rod in real time, accidents due to electrode rod damage can be prevented and time and cost can be significantly reduced.

[0070] FIG. 6 is a flowchart illustrating a method for detecting an abnormality of an electrode rod using an anomaly transformer model according to another embodiment, and FIG. 7 is a drawing illustrating an anomaly transformer model according to another embodiment.

[0071] Referring to FIG. 6, time series data collected by a data collector (110) and preprocessed to a predetermined size by a data preprocessor (120) is input to an anomaly determination device (130) (S210).

[0072] Referring to FIG. 6, the anomaly determination unit (130) can generate restored time series data corresponding to the input time series data and calculate a restoration error between the input time series data and the restored time series data (S220). The anomaly determination unit (130) can calculate the restoration error of the time series data for each electrode rod (2) included in the electric furnace (1). For example, when the electric furnace (1) includes three electrode rods (2), the anomaly determination unit (130) can calculate a restoration error of a different value for each electrode rod (2).

[0073] Referring to FIG. 7, the anomaly determiner (130) can generate restored time series data corresponding to the input time series data using an anomaly transformer model (132). Here, the anomaly transformer model (132) may be pre-trained based on normal time series data collected when the electrode rod (2) is not damaged. Therefore, when normal time series data with high similarity to each other are input, the anomaly transformer model (132) can restore time series data similar to the input, but when abnormal time series data with low similarity to the time series data used for pre-training is input, it cannot perform restoration similar to the input. In addition, when the anomaly transformer model (132) restores time series data that is not similar to the input, the restoration error between the input time series data and the restored time series data may increase.

[0074] In one embodiment, when a time series dataset of a predetermined size is input to the anomaly determiner (130), the anomaly determiner (130) can restore the time series dataset in parallel using the anomaly transformer model (132) and calculate the restoration error between the restored time series dataset and the input time series dataset at the same time. For example, the anomaly determiner (130) can perform restoration in parallel for 10 time series data within a time series dataset of size 10 collected for 0.2 seconds and determine 10 restoration errors between the restored time series data and the corresponding input time series data at the same time.

[0075] Referring to FIG. 6, the abnormality judgement device (130) can calculate the association discrepancy of input time series data (S230).

[0076] Referring to FIG. 7, the anomaly determiner (130) can calculate association inconsistencies using prior associations and series associations calculated in the anomaly attention layer of the anomaly transformer model (132). In the anomaly transformer model (132), it can be utilized that the anomalous time series data has a low correlation with the normal time series data, but a high correlation with the time points surrounding the anomalous time series data. On the other hand, the normal time series data has a high correlation with the entire time series data and also a high correlation with the surrounding time series data. The anomaly transformer model (132) can utilize the inconsistency in the associations of the anomalous time series data, and the mini-Max technique can be used to effectively calculate the association inconsistencies.

[0077] Referring to FIG. 6, the anomaly detector (130) can calculate an anomaly score of input time series data using the reconstruction error and associated inconsistency of the input time series data (S240). In one embodiment, the anomaly detector (130) can calculate the anomaly score by multiplying the reconstruction error and associated inconsistency of the input time series data.

[0078] Thereafter, the abnormality judgement device (130) can compare the abnormality score of the input time series data with a predetermined reference score and determine whether the electrode rod is damaged based on the comparison result (S250).

[0079] The reference score for the abnormal score may be determined in advance by considering the process progress, the average life of the electrode rod (2), etc. In one embodiment, if the abnormal score of the input time series data does not exceed the predetermined reference score, the abnormality determiner (130) may determine that no damage has occurred to the electrode rod (2), and if the abnormal score of the input time series data exceeds the predetermined reference score, the abnormality determiner (130) may determine that damage has occurred to the electrode rod (2). For example, the reference score may be determined as a value corresponding to the top 5% of the abnormal scores calculated based on normal learning data.

[0080] Alternatively, if the abnormality score of the input time series data exceeds a predetermined auxiliary reference score, the abnormality determiner (130) may preliminarily determine that damage to the electrode rod (2) is imminent. For example, the auxiliary reference score may be determined as a value corresponding to the top 8% of the abnormality scores calculated based on normal learning data.

[0081] When a time series data set of a predetermined size is input to an anomaly judger (130), the anomaly judger (130) can judge the occurrence of damage to the electrode rod (2) based on whether an anomaly score among the time series data in the time series data set exceeds a predetermined reference score.

[0082] In one embodiment, the abnormality determiner (130) can determine that damage has occurred to the electrode rod (2) when n or more consecutive time series data having an abnormality score greater than a reference score exist within a time series data set of a predetermined size, or m or more consecutive time series data having an abnormality score greater than a reference score exist. In this case, both m and n are natural numbers smaller than the size of the time series data set, and m may be greater than n.

[0083] For example, if six or more time series data having an abnormal score higher than the reference score are detected in a time series data set of size 10, or three or more are detected consecutively, the abnormality determiner (130) can determine that damage has occurred in the electrode rod (2).

[0084] Alternatively, the abnormality judger (130) may preliminarily judge that damage will soon occur to the electrode rod (2) if time series data having an abnormality score greater than an auxiliary reference score is detected within a time series data set of a predetermined size. At this time, the auxiliary reference score may be determined in advance by considering the process progress, the average lifespan of the electrode rod (2), etc., and having a size smaller than the reference score (for example, a value corresponding to the top 8% of the abnormality scores calculated based on normal learning data).

[0085] In one embodiment, the abnormality determiner (130) may preliminarily determine that damage will soon occur to the electrode rod (2) when n or more time series data having an abnormality score greater than an auxiliary reference score are continuously detected within a time series data set of a predetermined size, or m or more time series data having an abnormality score greater than an auxiliary reference score are detected. In this case, both m and n are natural numbers smaller than the size of the time series data set, and m may be greater than n.

[0086] For example, if six or more time series data having an abnormal score higher than the auxiliary reference score are detected in a time series dataset of size 10, or three or more are detected in succession, the abnormality determiner (130) can preliminarily determine that damage will soon occur to the electrode rod (2).

[0087] If the abnormal score of the input time series data exceeds a predetermined reference score, the abnormality determiner (130) may determine that the electrode rod (2) is damaged and report damage to the electrode rod (2) (S260). Alternatively, if the abnormal score of the input time series data exceeds a predetermined auxiliary reference score, the abnormality determiner (130) may preliminarily determine that the electrode rod (2) is about to be damaged and report a risk of damage to the electrode rod (2).

[0088] As described above, by utilizing the anomaly transformer model to detect abnormalities in the electrode rod in real time, accidents due to electrode rod damage can be prevented and time and cost can be significantly reduced.

[0089] Fig. 8 is a block diagram showing an abnormality detection device of an electrode rod according to one embodiment.

[0090] An anomaly detection device according to one embodiment may be implemented as a computer system, for example, a computer-readable medium. Referring to FIG. 8, a computer system (800) includes at least one processor (810) and a memory (820). The memory (820) is connected to the processor (810) and may store various information for driving the processor (810) or at least one program executed by the processor (810).

[0091] The processor (810) may implement the functions, processes, or methods proposed in the embodiments. The operation of the computer system (800) according to the embodiments may be implemented by the processor (810). At least one processor (810) may include at least one of a GPU, a CPU, and an NPU. When the operation of the computer system (800) is implemented by at least one processor (810), each task may be divided among the at least one processor (810) according to the load. For example, when one processor is a CPU, the other processor may be any one of a GPU, an NPU, an FPGA, and a DSP.

[0092] In the embodiments of the present disclosure, the memory (820) may be located internally or externally to the processor, and may be connected to the processor via various known means. The memory may be a variety of volatile or non-volatile storage media, and may include, for example, read-only memory (ROM) or random access memory (RAM).

[0093] Meanwhile, the embodiments are not implemented only through the devices and / or methods described so far, but may also be implemented through a program that realizes a function corresponding to the configuration of the embodiments or a recording medium on which the program is recorded. Such implementation can be easily implemented by a person skilled in the art to which the present disclosure pertains from the description of the embodiments described above. Specifically, the method according to the embodiments (e.g., an image preprocessing method, etc.) may be implemented in the form of program commands that can be executed through various computer means, and may be recorded on a computer-readable medium. The computer-readable medium may include program commands, data files, data structures, etc., singly or in combination. The program commands recorded on the computer-readable medium may be specially designed and configured for the embodiments, or may be known and usable by a person skilled in the art of computer software. The computer-readable recording medium may include a hardware device configured to store and execute the program commands. For example, the computer-readable recording medium may be magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, ROMs, RAMs, flash memories, etc. The program instructions may 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 through an interpreter, etc.

[0094] Although the embodiments have been described in detail above, the scope of the present disclosure is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concepts defined in the following claims also fall within the scope of the present disclosure.

Claims

1. A method for detecting an abnormality in an electrode rod of an electric furnace, A step of collecting time series data indicating the state of the electrode rod; A step of restoring the input time series data within the collected time series dataset using an autoencoder model; A step of calculating a restoration error between the input time series data and the restored time series data, and A step for determining whether the electrode rod is damaged based on the comparison result between the above restoration error and the reference value. How to include.

2. In paragraph 1, The step of collecting time series data indicating the status of the above electrode rod is: A step of collecting the time series data from a plurality of sensors used for process monitoring of the electric furnace, and A step of collecting position control values ​​used to control the position of the above electrode rod A method comprising:

3. In paragraph 1, The step of calculating the restoration error between the input time series data and the restored time series data is: A step of calculating the restoration error for each of the plurality of electrode rods included in the electric furnace. A method comprising:

4. In paragraph 1, A method in which the above autoencoder model is pre-trained to restore time series data similar to the normal time series data by using normal time series data collected when the electrode rod is not damaged as training data.

5. In paragraph 1, The above reference value is determined in advance by considering the process progress and the average lifespan of the electrode rod.

6. In paragraph 1, The step of determining whether the electrode rod is damaged based on the comparison result between the above restoration error and the reference value is as follows: A step of determining that damage has occurred in the electrode rod when n or more input time series data having a restoration error greater than the reference value are continuously detected within the time series data set. A method comprising: wherein n is a natural number smaller than the size of the time series dataset.

7. In paragraph 1, The step of determining whether the electrode rod is damaged based on the comparison result between the above restoration error and the reference value is as follows: A step of determining that damage has occurred in the electrode rod when m or more input time series data having a restoration error greater than the reference value are detected within the time series data set. A method comprising: wherein m is a natural number smaller than the size of the time series dataset.

8. In paragraph 1, A step of preliminarily determining that damage will soon occur to the electrode rod when input time series data having a restoration error greater than the auxiliary reference value is detected within the above time series data set. A method comprising: further including the auxiliary reference value, wherein the auxiliary reference value is smaller than the reference value.

9. A method for detecting abnormalities in electrode rods of an electric furnace, A step of collecting time series data indicating the state of the electrode rod; A step of restoring input time series data within a collected time series dataset and calculating a restoration error between the input time series data and the restored time series data, A step of calculating the correlation inconsistency of the above input time series data, A step of calculating an anomaly score of the input time series data based on the above restoration error and the above association mismatch, and A step of determining whether the electrode rod is damaged based on the comparison result between the above-mentioned abnormal score and the reference score. How to include.

10. In paragraph 9, The step of restoring the above input time series data and calculating the restoration error between the input time series data and the restored time series data is: Step of dividing the collected time series data into predetermined sizes and inputting them into the anomaly transformer model. A method comprising:

11. In paragraph 10, The step of calculating the correlation inconsistency of the above input time series data is: A step of calculating the association mismatch using the prior association and series association calculated in the anomaly attention layer of the anomaly transformer. A method comprising:

12. In paragraph 10, A method in which the above anomaly transformer model is pre-trained to restore time series data similar to the normal time series data by using normal time series data collected when the electrode rod is not damaged as training data.

13. In paragraph 10, The step of calculating the anomaly score of the input time series data based on the above restoration error and the above association mismatch is as follows: A step of calculating the product of the above restoration error and the above association mismatch as the above abnormality score. A method comprising:

14. In paragraph 9, The above reference score is determined in advance by considering the process progress and the average lifespan of the electrode rod.

15. In paragraph 9, The step of determining whether the electrode rod is damaged based on the comparison result between the above-mentioned abnormal score and the reference score is as follows: A step of determining that damage has occurred in the electrode rod when n or more input time series data having an abnormal score greater than the reference score are continuously detected within the time series data set. A method comprising: wherein n is a natural number smaller than the size of the time series dataset.

16. In paragraph 9, The step of determining whether the electrode rod is damaged based on the comparison result between the above-mentioned abnormal score and the reference score is as follows: A step of determining that damage has occurred in the electrode rod when m or more input time series data having an abnormal score greater than the reference score are detected within the time series data set. A method comprising: wherein m is a natural number smaller than the size of the time series dataset.

17. In paragraph 9, A step of preliminarily determining that damage will soon occur to the electrode rod when input time series data having an abnormal score greater than the auxiliary reference score is detected within the above time series data set. A method further comprising: wherein the auxiliary reference score is smaller than the reference score.

18. A device for detecting abnormalities in the electrode rod of an electric furnace, A processor and a memory, wherein the memory stores instructions that cause the processor to execute a plurality of steps, wherein the plurality of steps include: A step of collecting time series data from multiple sensors used for process monitoring of the above electric furnace, A step of restoring input time series data with a predetermined size using an artificial neural network model based on a restoration method, A step of calculating a restoration error between the input time series data and the restored time series data, and A step for determining whether the electrode rod is damaged based on the above restoration error. A device comprising:

19. In paragraph 18, The artificial neural network model of the above restoration-based method is an autoencoder model, The step of determining whether the electrode rod is damaged based on the above restoration error is as follows: A step for determining that the electrode rod is damaged when the above restoration error is greater than the reference value. A device comprising 20. In paragraph 18, The artificial neural network model of the above restoration-based method is an anomaly transformer model, and the multiple steps are A step of calculating the correlation inconsistency of the above input time series data, and A step of calculating an anomaly score of the input time series data based on the above restoration error and the above association mismatch. A device further comprising:

21. In paragraph 20, The step of determining whether the electrode rod is damaged based on the above restoration error is as follows: A step of determining whether the electrode rod is damaged based on the comparison result between the above-mentioned abnormal score and the reference score. A device comprising:

22. In paragraph 18, A device wherein the plurality of sensors include a current sensor for measuring current supplied to the electrode rod, a voltage sensor for measuring voltage formed on the electrode rod, and an impedance sensor for measuring impedance of the electrode rod.

Images