Sensor abnormal value detection system and method using integrated algorithm taking feature variables into consideration

Abstract

The present invention relates to a system for detecting an abnormal value of a sensor, which uses an integrated algorithm that considers a feature variable, and which can classify the degree of inclusion of a false abnormal value signal due to noise by cumulatively analyzing various sensing information of a monitoring target such as a rotating body in time domain. And by applying an abnormal value detection algorithm which is optimized according to noise (characteristic variables) contained in each time domain, the accuracy of abnormal value detection can be improved, and waste of resources caused by algorithm calculation can be reduced to the greatest extent. The system can comprise a sensor and a management server, wherein the management server comprises a time analysis unit, a waveform analysis unit, a determination unit and a detection unit.

Description

Technical Field

[0001] The present invention relates to a sensor outlier detection technology using an ensemble algorithm considering feature variables. More specifically, the present invention relates to a sensor outlier detection system and method using an ensemble algorithm considering feature variables, which can classify the inclusion degree of false outlier signals caused by noise by performing time-domain cumulative analysis on various sensing information of a monitoring target such as a rotating body, and can improve the accuracy of outlier detection by applying an outlier detection algorithm optimized according to the noise (feature variables) included in each time domain, and can minimize the waste of resources caused by algorithm calculation at the same time. Background Art

[0002] Water pipes for water supply in houses, stores, factories, apartments, etc. are buried underground, and multiple water pipes are connected like a network to form a water pipe network. If old water pipes are not replaced in time, not only will a large amount of water be wasted, but also the likelihood of a leakage accident will increase due to abnormal states occurring in the pipes, such as pipe cracking, pipe inclination, foreign matter accumulation in the pipe, and an increase in the flow velocity in the pipe. Therefore, prior maintenance is required before a leakage accident occurs.

[0003] In addition, industrial machine tools are a general term for machines used for manufacturing equipment, and are machines that use cutting tools and process metals and other materials into the required shapes while generating chip fragments by cutting, turning, boring, drilling, threading, grinding, and other methods on metals.

[0004] Industrial machine tools that perform cutting on a machining target by rotation inevitably generate vibrations. In particular, when abnormal vibrations occur in the vibrations generated during the cutting of industrial machine tools, the usage efficiency of the tools and the accuracy of the workpieces (low quality) may be reduced. Specifically, abnormal vibrations can cause various effects, such as tool wear and breakage, spindle bearing wear and failure, poor surface roughness of the workpiece, low product quality, and increased energy consumption.

[0005] Therefore, in the past, in order to monitor abnormal states of targets such as pipes, buildings, or industrial machine tools based on rotating bodies, vibration sensors were usually installed on the rotating bodies of pipes, buildings, or factories, and accident locations were detected through the vibrations generated by the targets.

[0006] However, in addition to the signals detected solely from objects such as pipelines, rotating bodies, and buildings, a large amount of noise is introduced into the detection signals of the sensors. For example, vehicle operations and traffic voice prompts around the objects, construction noise around the objects, changes in the flow rate inside the pipeline, etc. Therefore, in addition to the true outlier signals caused by abnormal states such as changes in the flow rate inside the pipeline, there are also false outlier signals caused by various reasons, resulting in the drawback that the detection of abnormal states frequently makes mistakes.

[0007] Korean Patent No. 10-2269144 discloses a technique for extracting industrial machine operation data including normal data and abnormal data based on the vibrations generated during the operation of an industrial machine, and determining the multi-dimensional abnormal directionality of the industrial machine based on the operation data.

[0008] However, in addition to the signals detected solely from the rotating body (i.e., the industrial machine), a large amount of surrounding noise is introduced into the detection signals of the sensor. Therefore, in addition to the true outlier signals caused by the abnormal state of the rotating body, there are also a large number of false outlier signals caused by various reasons, resulting in the drawback that the detection of the abnormal state of the rotating body frequently makes mistakes.

[0009] Therefore, it is necessary to use a sophisticated outlier detection algorithm to be able to more accurately distinguish the true outlier signals as the target from the false outlier signals.

[0010] Patent Document 1: Korean Patent Gazette No. 10-2269144 (June 18, 2021) Summary of the Invention

[0011] Technical Problem to be Solved by the Invention

[0012] The problem to be solved by the present invention is to provide an outlier detection system and method that classifies the inclusion degree of false outlier signals according to noise by cumulatively analyzing various sensing information of the monitoring target in the time domain, and variably selects and applies an outlier detection algorithm in each time domain.

[0013] Technical Means for Solving the Problem

[0014] The present invention proposes a sensor outlier detection system using an integrated algorithm considering characteristic variables as one of the methods for solving the above problems.

[0015] A sensor outlier detection system according to an embodiment includes: a vibration sensor and a supervisory server. The vibration sensor is disposed on a monitoring target. The supervisory server includes: a time analyzer that cumulatively stores detection signals of the sensor and differentiates a plurality of time domains having signal similarity within a set standard by analyzing changes in waveforms of the detection signals over time on a daily basis; a waveform analyzer that confirms characteristic variables by analyzing the detection signals in each of the plurality of time domains; a decision unit that integratively learns a plurality of outlier detection algorithms based on the detection signal values in each time domain and the characteristic variables confirmed in each time domain, and selects an optimal outlier detection algorithm for each time domain; and a detector that determines an abnormal state of the target by applying a preset outlier detection algorithm to each of the plurality of time domains.

[0016] In one embodiment, the monitoring target may also be one of a water pipe, an industrial machine tool, a steam turbine, a gas turbine, a wind turbine, a propeller, and an airscrew operating in a power plant.

[0017] In one embodiment, the time analyzer converts the detection signals into a frequency domain, and the waveform analyzer may also confirm the characteristic variables by analyzing the composition of frequencies included in the detection signals in the converted frequency domain.

[0018] In one embodiment, the decision unit may also determine other outlier detection algorithms by considering at least one of the number or intensity of frequencies of the characteristic variables included in the detection signals in the frequency domain.

[0019] In one embodiment, the characteristic variables may also include at least one of vibrations generated by other mechanical equipment around the monitoring target during initial operation, vibrations generated by continuous operation of the other mechanical equipment, vibrations according to the output of the other mechanical equipment, and vibrations of vehicles passing by in the vicinity.

[0020] In one embodiment, the detector may not perform a determination of an abnormal state in a time domain where changes in the waveforms in the plurality of time domains continuously fall below a standard value.

[0021] The present invention proposes a sensor outlier detection method using an integrated algorithm considering characteristic variables as another method among methods for solving the above problems.

[0022] The method for detecting sensor outliers in an embodiment includes: the step of the time analysis unit of the monitoring device cumulatively storing the vibration detection signals of the sensors set at the target object; the step of the time analysis unit analyzing the change in the waveform of the detection signals over time in days and dividing the time periods with signal similarity within the set standard into several; the step of the waveform analysis unit of the monitoring device confirming the characteristic variables by analyzing the detection signals in each of the multiple time domains; the step of the determination unit of the monitoring device integrating and learning multiple outlier detection algorithms based on the detection signal values in each time domain and the characteristic variables confirmed in each time domain, and determining the optimal outlier detection algorithm for each time domain; and the step of the detection unit of the monitoring device judging the abnormal state of the target object by applying a preset outlier detection algorithm to each of the multiple time domains.

[0023] In one embodiment, the monitored target object can also be any one of a water pipe, an industrial machine tool, a steam turbine, a gas turbine, a wind turbine, a propeller, and an airscrew operating in a power plant.

[0024] In one embodiment, the time analysis unit converts the detection signals into the frequency domain, and the waveform analysis unit can also confirm the characteristic variables by analyzing the composition of the frequencies included in the converted detection signals in the frequency domain.

[0025] In one embodiment, the determination unit can also determine other outlier detection algorithms by considering at least one of the number or intensity of the frequencies of the characteristic variables included in the detection signals in the frequency domain.

[0026] In one embodiment, the characteristic variables can also include at least one of the vibration or noise generated by other mechanical equipment around the monitored target object during initial operation, the vibration or noise generated due to the continuous operation of the other mechanical equipment, and the vibration or noise according to the output of the other mechanical equipment.

[0027] In one embodiment, the detection unit may not perform the judgment of the abnormal state in the time domains where the change in the waveform in multiple time domains continuously falls below the standard value.

[0028] Advantages of the Invention

[0029] According to the embodiments of the present invention, by applying different outlier detection algorithms optimized according to noise (characteristic variables) in each time domain, not only can the accuracy of outlier detection be improved, but also the waste of resources caused by algorithm calculation can be minimized.

[0030] According to an embodiment of the present invention, in order to be able to more accurately distinguish an outlier signal from a false outlier signal, an accurate outlier detection algorithm is used, so that the abnormal state of the monitoring target can be known in advance, enabling preventive measures to be taken, thereby preventing the occurrence of safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a structural diagram of a sensor outlier detection system using an integrated algorithm considering characteristic variables according to Embodiment 1.

[0032] Figure 2 shows what is included in Figure 1 a block diagram of the detailed configuration of the management server in the system of.

[0033] Figure 3 shows Figure 2 a conceptual diagram of the detailed operation of the determination unit in the management server of.

[0034] Figure 4 is a diagram showing the detailed configuration of a sensor outlier detection system according to Embodiment 2 and a monitoring device 300 included in the system.

[0035] Figure 5 is a flowchart showing a sensor outlier detection method according to Embodiment 3.

[0036] DESCRIPTION OF REFERENCE NUMERALS

[0037] 100: Vibration sensor; 200: Management server; 210: Time analysis unit; 220: Waveform analysis unit; 230: Determination unit; 240: Detection unit. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] Several embodiments of the present invention will be described in detail below with reference to the drawings. However, it should be understood that this is not intended to limit the present invention to any specific embodiment, and all transformations, equivalents, and substitutions including the technical idea of the present invention are included within the scope of the present invention.

[0039] In this specification, unless the context clearly dictates otherwise, expressions in the singular form include expressions in the plural form.

[0040] In this specification, unless explicitly described to the contrary, when describing that a certain component "has" or "comprises" a certain sub-component, this means that other components are not excluded, but other components may also be included.

[0041] In this specification, the terms “... Unit”, “... Module”, and “Component” refer to a unit that processes at least one function or operation, and can be implemented by hardware, software, or a combination of hardware and software.

[0042] In this specification, the description of “connect” can refer to a direct connection between two components, but is not limited to this. It can also mean a connection through one or more other components placed between the components.

[0043] Figure 1 It is a structural diagram of a sensor outlier detection system using an integrated algorithm considering characteristic variables according to Embodiment 1.

[0044] As Figure 1 shown, the vibration sensor outlier detection system of Embodiment 1 may include: a vibration sensor 100 and a management server 200.

[0045] The vibration sensor 100 is disposed on the monitoring target and senses the vibration generated by the monitoring target.

[0046] In an embodiment of the present invention, the monitoring target may refer to a rotating body.

[0047] In an embodiment of the present invention, the rotating body may refer to rotating equipment such as industrial machine tools, steam turbines, and gas turbines operating in a power plant, or rotor-based rotating bodies such as wind turbines, propellers, and airscrews.

[0048] The vibration sensor 100 can be directly disposed on the monitoring target or indirectly disposed in the equipment room accommodating the monitoring target. For example, when the monitoring target is a turbine, the vibration sensor 100 can be disposed on the outer peripheral surface of the turbine blade or indirectly disposed on the inner side surface of the equipment room around the turbine.

[0049] On the other hand, the sensed information measured by the vibration sensor 100 can be transmitted to the management server 200 in real time through a wired or wireless wide-area communication network, or can be accumulated and stored in a storage device (not shown) connected to the vibration sensor 100 at a very short distance for a specified time, and then transmitted through the wide-area communication network at a preset cycle. In addition, it can also be accumulated and stored in the storage device, and then collected by the mobile terminal of the manager close to the vibration sensor 100 at a very short distance, and then indirectly transmitted to the management server 200.

[0050] The supervisory server 200 detects an abnormal state of a monitoring target by analyzing the sensing information collected from the vibration sensor 100.

[0051] Specifically, the supervisory server 200 analyzes the characteristics of the detection signals in each time domain based on the sensing information collected from the vibration sensor 100, differentiates the time domains maintaining similar characteristics, confirms the characteristic variables of each time domain, and detects the abnormal values of the monitoring target in a manner of an integrated algorithm that applies different strategies for each time domain based on the detection signal values and the characteristic variables of each time domain. As a reference, the integrated strategy herein refers to determining an optimal outlier detection algorithm among multiple outlier detection algorithms.

[0052] Outlier detection refers to finding signal individuals with patterns different from the expected ones in the data, and can also refer to creating a model for finding heterogeneous data with characteristics different from the existing data based on the learned data.

[0053] For example, the heterogeneous data with characteristics different from the existing data can also be the frequencies caused by cracks in the blades, the frequencies caused by the blades being bent due to external factors, and the frequencies caused by foreign objects adhering to the surface of the blades.

[0054] Hereinafter, with reference to Figure 2 a detailed description will be given of the detailed configuration of the supervisory server 200.

[0055] Figure 2 is a block diagram showing the detailed structure of the supervisory server 200 included in the Figure 1 system.

[0056] As Figure 2 shown, the supervisory server 200 may include: a time analysis unit 210, a waveform analysis unit 220, a determination unit 230, and a detection unit 240.

[0057] The time analysis unit 210 (time analyzer) accumulatively stores the detection signals of the vibration sensor 100, analyzes the change in the waveform of the detection signals over time on a daily basis, and differentiates several time domains with signal similarity within a set standard.

[0058] The time analysis unit 210 converts the detection signals of the vibration sensor 100 into digital signals in the time domain, and converts the converted digital detection signals back into the frequency domain.

[0059] At this time, the time analysis unit 210 can also use Fourier Transform or Fast Fourier Transform to convert the digital detection signal in the time domain into the frequency domain, and then analyze the waveform changes on a daily basis, and distinguish several time domains with signal similarity within the set standard.

[0060] Among them, the time domain can refer to n time units (n is a positive real number), or a day can be roughly divided into two to distinguish the daytime time domain and the nighttime time domain, or it can be divided into the commuting time domain, the daytime working time domain, and the nighttime time domain. However, this is only an example. As long as it is a time domain with signal similarity within the preset standard, a day can be divided into m (m is a positive integer) detailed time domains.

[0061] As a specific example, assume that the vibration sensor 100 detects the vibration of a rotating device (i.e., a rotating body) in a factory with various mechanical equipment. Among the mechanical equipment in the factory, some equipment operates continuously for 24 hours regardless of whether employees are present, and some equipment can only operate when employees are working.

[0062] In this case, since the machine starts running during the operation time of the factory's mechanical equipment (from 07:00 am to 09:30 am), the vibration signal generated by the machine during its initial operation (for example, engine startup vibration) will be mixed with the vehicle vibration signal of vehicles passing by (for example, commuting vehicles) when the factory is located near a road. Therefore, the detection signal in the frequency domain of this time domain will show considerable complexity (or signal changes).

[0063] During the daytime working hours (from 09:31 am to 04:30 pm), the noise of the mechanical equipment during its initial operation is relatively low, the mechanical equipment operates continuously, and there are fewer vehicles passing by, so the complexity (or signal changes) of the detection signal is relatively low compared to the operation time domain of the factory's mechanical equipment.

[0064] From 04:31 pm, when some mechanical equipment stops operating, to 06:59 am the next day, some mechanical equipment does not operate, so it has the lowest signal complexity (or signal changes) compared to the operation time domain of the factory's mechanical equipment and the daytime working time domain.

[0065] Therefore, as long as the signal analysis unit analyzes the detection signals accumulated by the rotating body in the factory on a daily basis, it can roughly confirm whether the complexity characteristics of the detection signals in the operation time domain of the mechanical equipment, the daytime working time domain, and the commuting time domain when some mechanical equipment stops operating are maintained similar. Therefore, the signal analysis unit divides a day of the rotating body in the factory into three time domains.

[0066] The above assumptions are characteristics of the time domain that can generally be predicted. However, within a factory, the characteristics of the detection signals measured in each detailed area may also vary significantly within each time domain.

[0067] Therefore, the signal analysis unit analyzes the sensing information cumulatively stored by the vibration sensor 100 at a specific location during a preset period (such as one month, one quarter, or half a year), and analyzes the change in the waveform of the detection signal over time on a daily basis to divide the time domains with signal similarity within the set standard into several. Each time domain distinguished by the signal analysis unit can be automatically changed through periodic analysis.

[0068] On the other hand, if the detection signal collected by the vibration sensor 100 is an abnormal signal limited to the monitoring target, then this abnormal signal will only consist of signals in the normal state and signals in the abnormal state. Therefore, in principle, even if a vibration sensor abnormal value detection algorithm with the same combination is integrated regardless of the time domain of the detection signal, there will be no significant difference in the detection accuracy or the use of system resources for detection.

[0069] As a specific example, the vibration sensor 100 provided on the rotating body measures the vibration generated when the blades of the rotating body rotate. Detection signals can be collected in the normal state where there are no special problems with the blades of the rotating body, or in an abnormal state (for example, when there are cracks in the blades of the rotating body, the blades of the rotating body are bent due to external factors, or there is sediment attached to the blades of the rotating body).

[0070] However, in reality, in addition to the signals of the simple rotating body in the above normal and abnormal states, vibrations or noises generated by various other situations are also introduced (for example, vibrations or noises generated when other mechanical equipment around the rotating body is initially operating, vibrations or noises generated due to the continuous operation of other mechanical equipment, vibrations or noises according to the output of other mechanical equipment used, vibrations or noises of vehicles passing by in the vicinity), resulting in a complex waveform of the detection signal.

[0071] In other words, in addition to the true abnormal value signals according to the abnormal state of the rotating body, there are also many false abnormal value signals generated for various reasons, causing frequent errors in detecting the abnormal state of the rotating body. Therefore, the present invention applies an integrated algorithm to improve the accuracy of detecting true abnormal values.

[0072] However, that is not all. It is necessary to consider that the vibrations or noises generated by other mechanical equipment around the rotating body during initial operation, the vibrations or noises generated by the continuous operation of other mechanical equipment, the vibrations or noises according to the output of other mechanical equipment used, and the vibrations or noises of vehicles passing by in the vicinity occur in different time domains throughout the day. Additionally, it is necessary to consider that, depending on the placement position of the monitoring target (i.e., the rotating body), the noises generated in each time domain will also vary.

[0073] For example, in time domains such as early morning or midnight, there is little vehicle movement, the usage of drinking water is low, and the projects that should be carried out during the day are interrupted. Therefore, the possibility of including fewer false outlier signals increases. Even in the daytime time domain, the types of noises generated in each detailed interval will also be different.

[0074] Therefore, in order to detect outliers in the vibration sensor, the present invention adopts an integrated algorithm and proposes a solution that applies an optimal integration strategy to each time domain that maintains similar signal characteristics on the premise that the characteristics of the detection signals in each time domain of a day are assumed to be different.

[0075] For this purpose, the present embodiment is characterized in that the waveform analysis unit 220 confirms characteristic variables by analyzing the detection signals in each distinguished time domain, the determination unit 230 determines the optimal outlier detection algorithm for judging the abnormal state of the target by integrating the outputs of multiple outlier detection algorithms, and differently determines the outlier detection algorithm for the integration target according to each distinguished time domain and the confirmed characteristic variables.

[0076] The waveform analysis unit 220 can confirm characteristic variables by analyzing the composition of the frequencies included in the detection signals in the frequency domain converted by the time analysis unit 210.

[0077] The characteristic variables may include: the vibrations or noises generated by other mechanical equipment around the monitoring target (i.e., the rotating body) during initial operation, the vibrations or noises generated by the continuous operation of other said mechanical equipment, the vibrations or noises according to the output of other said mechanical equipment used, and the vibrations or noises of vehicles passing by in the vicinity.

[0078] In addition, when the monitoring target is an industrial machine tool, the characteristic variables may further include the vibrations generated by the machining target. Because, even when using the same rotating body (e.g., a milling machine) for cutting, the vibrations generated will vary depending on the machining target.

[0079] The composition of the frequencies included in the digital detection signal in the frequency domain may refer to separating the frequencies with different intensities (Hz) converted from the vibration signal generated by the rotating body and confirming the number of each separated frequency. If there are three frequencies with different intensities (Hz), the digital detection signal in the corresponding frequency domain includes three vibration signals.

[0080] For example, when the rotating body operates in a factory with various types of equipment, the frequencies generated due to the operation of other surrounding equipment or the frequencies generated due to indirect causes occurring in the vicinity, such as the vibration signal of a vehicle passing by in the vicinity, may also have different frequency intensities (Hz) respectively, and the vibration signals with different frequency intensities can be confirmed as characteristic variables.

[0081] Figure 3 It shows Figure 2 a conceptual diagram of the detailed operation of the determination unit 230 in the management server 200.

[0082] As Figure 3 shown, the determination unit 230 (decision) integrates and learns multiple outlier detection algorithms based on the detection signal values in each time domain and the characteristic variables in each time domain to select the optimal outlier detection algorithm for each time domain. Among them, the detection signal value can be a digital detection signal value or a signal value in the frequency domain.

[0083] For example, as described above, according to the operating time domain of the mechanical equipment, the daytime working time domain, and the off-duty time domain when some of the mechanical equipment stops operating, which are distinguished by the time analysis unit 210, the characteristic variables will be constituted differently respectively.

[0084] The determination unit 230 considers the time of each area and the frequency intensity or the number of frequencies of the characteristic variables constituted at the time of each area, and selects the optimal outlier detection algorithm from multiple outlier detection algorithms according to the signal complexity caused by the detection signal of the simple rotating body and the characteristic variables at each area time.

[0085] For example, the determination unit 230 basically determines different outlier detection algorithms for each time domain, but can determine other outlier detection algorithms by further considering the number of frequencies of the characteristic variables included in the detection signal in the frequency domain included in the time domain, so the decision-making ability of the algorithm can be improved.

[0086] In addition, even if the composition of the frequencies of the characteristic variables included in the detection signal in the frequency domain is the same as three, the determination unit 230 can determine other outlier detection algorithms according to the intensity of each frequency, so the decision-making ability of the algorithm can be improved.

[0087] On the other hand, the determination unit 230 may use the voting method, bagging algorithm, boosting algorithm, or stacking method in ensemble learning. The voting method is a method of predicting the final result by voting on the results of various algorithms.

[0088] The outlier detection algorithm used by the determination unit 230 may use at least one of the interquartile range (IQR), z-score, generalized extreme studentized deviate test (Generalized ESD test), isolation forest, local outlier factor (LOF), and one-class support vector machine (One-Class SVM).

[0089] The determination unit 230 may use any one of the clustering basis outlier detection algorithm, classification basis outlier detection algorithm, and statistical basis outlier detection algorithm, and thus is not limited to the above-mentioned interquartile range, etc.

[0090] The interquartile range is defined as the distance between the first quartile (25%) and the third quartile (75%) after arranging the data in ascending order. Moreover, if the data value is greater than or less than the value of the interquartile range multiplied by 1.5 plus or minus the median, it is determined as an outlier.

[0091] The z-score is also known as the standardized score. Under the assumption that the data follows a normal probability distribution, considering the standard deviation, it is determined whether it is an outlier by looking at the distance between the data and the mean.

[0092] The generalized extreme studentized deviate test uses the same statistic as the Grubbs test for verification, but it is a method to improve the shortcomings of the Grubbs verification. For reference, the Grubbs test (verification) has the absolute deviation of the data value that is farthest from the mean, and verifies whether it is an outlier under the assumption that the data is from a normal distribution.

[0093] The isolation forest utilizes the fact that outliers are partitioned less than normal data, partitions the data into arbitrary partitions, and thereby determines outliers.

[0094] The determination unit 230 applies an integration strategy differently based on the result values of each outlier detection algorithm and based on the detection signal values and characteristic variables of each time domain.

[0095] As a reference, the integration strategy mentioned above refers to determining an optimal outlier detection algorithm among multiple outlier detection algorithms by using a voting method.

[0096] According to this integration strategy, when the time domain is divided into a commuting time domain T1, a daytime working time domain T2, and a nighttime time domain T3, integration strategy 1 (IQR) can be determined in the commuting time domain T1, integration strategy 2 (z-score) can be determined in the daytime working time domain T2, and integration strategy 3 (Generalized ESD test) can be determined in the nighttime time domain T3.

[0097] As described above, by applying different outlier detection algorithms optimized according to noise (characteristic variables) in each time domain, not only can the accuracy of outlier detection be improved, but also the waste of resources caused by algorithm calculation can be minimized.

[0098] The detection unit 240 (detector) determines the abnormal state of the target object by applying a preset outlier detection algorithm to each of the multiple time domains.

[0099] For example, the detection unit 240 may not perform the determination of the abnormal state in a time domain where the change in the waveform in multiple time domains continuously falls below the standard value. This can also minimize the waste of resources caused by algorithm calculation.

[0100] On the other hand, in this embodiment, the monitored target object may include a pipeline in addition to a rotating body.

[0101] The pipeline may refer to a water pipe, a sewer pipe, or a liquefied natural gas pipe buried underground or installed in a building, may also refer to a water pipe installed in a valve chamber, or may also refer to a conveying pipe through which hydrogen or gaseous or liquid fossil fuels pass. However, the monitored target object is not necessarily limited to these examples and should be understood to include the concept of any tubular member with fluid flowing inside.

[0102] A building is a general term for houses, buildings, bridges, towers, or their support frames, etc. A rotating body refers to a rotor-based rotating member such as a steam turbine, a gas turbine, a wind turbine, a propeller, or an airscrew.

[0103] In addition, in this embodiment, the vibration sensor 100 (sensor) can be disposed on the monitoring target object, and can sense sounds in addition to vibrations. The vibration sensor 100 (sensor) can be a separate sensor that serves as a vibration sensor or a sound sensor, or can be an integrated sensor that simultaneously measures vibrations and sounds.

[0104] <Embodiment 2>

[0105] Embodiment 2 relates to a technique for detecting outliers by sending sensed information measured from a monitoring target object to a monitoring device 300 at a very short distance instead of a management server at a long distance, and determining an optimal outlier detection algorithm among multiple outlier detection algorithms by considering different characteristic variables in each time domain.

[0106] Among them, the monitoring device 300 at a very short distance can refer to a single device connected to the sensor 310 through short-range wireless communication or wired communication, or can refer to a module connected to the inside of the sensor 310 through a circuit. When implemented as a module inside the sensor 310, the sensor 310 can refer to an edge sensor 310 that independently executes an outlier detection algorithm.

[0107] Figure 4 FIG. is a diagram showing a detailed configuration of a sensor outlier detection system according to Embodiment 2 and a monitoring device 300 included in the system.

[0108] As shown in Figure 4 FIG., a monitoring device 300 is disposed at a very short distance from the monitoring target object. The monitoring device 300 receives a detection signal from the monitoring target object and analyzes the received detection signal to determine an abnormal state. As a reference, the very short distance from the monitoring target object can refer to the inside of an equipment room where the monitoring target object is disposed, or can refer to the inside of a building.

[0109] The monitoring device 300 includes: a sensor 310, a time analysis unit 320, a waveform analysis unit 330, a determination unit 340, and a detection unit 350.

[0110] The only difference is that the sensor 310 transmits the sensed information to the time analysis unit 320 connected through short-range wireless communication, wired communication, or a bus on a circuit, instead of transmitting it to a management center through a wide-area communication network or via a manager's mobile terminal. Other technical structures are the same as those of the sensor 310 in Embodiment 1, so repeated descriptions are omitted.

[0111] The signal analysis unit accumulatively stores the detection signals of the sensor 310 in a storage device (not shown), and divides the time domain with signal similarity within a set standard into several by analyzing the change of the waveform of the detection signals over time on a daily basis. As described above, the only difference is that the time analysis unit 320 receives the detection signals from the sensor 310 connected through short-range wireless communication, wired communication, or a data bus on the circuit, and the other technical structures are the same as those of the analysis unit in Embodiment 1, so the repeated description is omitted.

[0112] The technical structures of the waveform analysis unit 330, the determination unit 340, and the detection unit 350 are the same as those of the waveform analysis unit 330, the determination unit 340, and the detection unit 350 in Embodiment 1, so the repeated description is omitted.

[0113] However, the detection unit 350 in Embodiment 2 may also have the following function: when an outlier of the sensor 310 is found in the monitored target due to execution of integration, it alerts the remote management device 400 of the occurrence or the possibility of the occurrence of an abnormal state in real time or at a preset time through a wide-area communication network.

[0114] In addition, the detection unit 350 in Embodiment 2 may not perform the judgment of the abnormal state in the time domain where the change of the waveform in multiple time domains continuously falls below the standard value, and thus may not send any alert to the management device 400 either.

[0115] <Embodiment 3>

[0116] Embodiment 3 relates to a method for detecting outliers of sensors executed in the monitoring device of Embodiment 2.

[0117] Figure 5 It is a flowchart showing the method for detecting outliers of sensors according to Embodiment 3.

[0118] As Figure 5 shown, the method for detecting outliers of sensors in Embodiment 3 may include: a step S110 of accumulatively storing detection signals; a step S120 of dividing the time periods with signal similarity into multiple; a step S130 of confirming characteristic variables; a step S140 of determining an outlier detection algorithm; and a step S150 of judging the abnormal state of the target.

[0119] In the step S110 of accumulatively storing detection signals, the time analysis unit of the monitoring device accumulatively stores the detection signals of the sensor.

[0120] In the step S120 of dividing the time periods with signal similarity into multiple, the time analysis unit analyzes the change of the waveform of the detection signals over time on a daily basis to divide the time domain with signal similarity within a set standard into several.

[0121] The time analysis unit converts the detection signal of the vibration sensor into a digital signal in the time domain, and then converts the converted digital detection signal into the frequency domain again.

[0122] At this time, the time analysis unit can also use the Fourier Transform or the Fast Fourier Transform to convert the digital detection signal in the time domain into the frequency domain, and then analyze the waveform changes on a daily basis, and divide the time domain with signal similarity within the set standard into several.

[0123] In the step S130 of confirming the characteristic variables, the waveform analysis unit of the monitoring device determines the characteristic variables by analyzing the composition of the frequencies included in the detection signal in the frequency domain converted in the time analysis unit.

[0124] In the step S140 of determining the outlier detection algorithm, the determination unit of the monitoring device integrates and learns multiple outlier detection algorithms based on the detection signal values of each time domain and the characteristic variables of each time domain, and selects any one outlier detection algorithm for each time domain.

[0125] In the step S150 of judging the abnormal state of the target object, the detection unit of the monitoring device judges the abnormal state of the target object by applying a preset outlier detection algorithm to each of the multiple time domains.

[0126] For example, in the time domain where the waveform changes in multiple time domains continuously fall below the standard value, the detection unit may not perform the judgment of the abnormal state, so it can also minimize the waste of resources caused by algorithm calculation.

[0127] In addition, in the time domain where the waveform changes in multiple time domains continuously fall below the standard value, the detection unit may not perform the judgment of the abnormal state, and at this time, no alarm may be sent to the management server.

[0128] As mentioned above, several embodiments of the present invention have been described. It is easy for those skilled in the art with ordinary knowledge in this technical field to understand that various modifications and changes can be made to the present invention without departing from the spirit and scope of the present invention described in the appended claims.

[0129] In addition, the invention of the method in the above embodiments can be implemented by a program or a computer-readable recording medium storing the program.

[0130] That is, the present invention can be implemented in the form of an application program, or as a software program that can be executed on mobile terminals such as smart phones and tablet computers that run on Android of Google or iOS of Apple, or as a software program that can be executed on wearable devices such as Google Glass, Apple Watch, Samsung Galaxy Watch, smart watches, etc., or as a software program that can be executed on laptops, desktop computers, etc. that run on Windows of Microsoft or ChromeOS of Google.

[0131] In addition, some functions of the above-mentioned device or system are implemented in a practical manner by a program of instructions for implementing the function, and thus are included in and provided on a computer-readable recording medium. The computer-readable recording medium may include individual program instructions, data files, data structures, etc. or combinations thereof. Examples of the computer-readable recording medium include: magnetic media (e.g., hard disks, floppy disks, and magnetic tapes); optical recording media (e.g., CD-ROMs, DVDs); magneto-optical media (e.g., floptical disks); and hardware devices specifically configured to store and execute program instructions (e.g., ROMs, RAMs, flash memories, USB memories, etc.).

[0132] This invention is supported by the following Korean national research and development project.

[0133] Subject matter unique number: 1425174611

[0134] Subject matter number: S3215463

[0135] Department name: Small and Medium-sized Venture Business Department

[0136] Name of the subject management (specialized) agency: Small and Medium-sized Enterprise Technology Information Promotion Agency

[0137] Name of the research project: Empirical project for the development of a fault prediction and diagnosis system through vibration variability analysis of rotating equipment in an IoT-based smart factory

[0138] Name of the research subject: Empirical project for the development of a fault prediction and diagnosis system through vibration variability analysis of rotating equipment in an IoT-based smart factory

[0139] Contribution rate: 1 / 1

[0140] Name of the subject executing agency: Eel Technology Co., Ltd.

[0141] Research period: April 1, 2022 to March 31, 2024.

Claims

1. A sensor outlier detection system using an integrated algorithm taking into account characteristic variables, characterized in that: It includes a vibration sensor and a management server, wherein the vibration sensor is arranged on the monitoring target object; The management server comprises: a time analysis unit that accumulates and stores the detection signal of the vibration sensor, analyzes the change of the waveform of the detection signal according to time in units of days, and distinguishes a plurality of time domains having signal similarity within a set standard, a waveform analysis unit, which confirms a characteristic variable by analyzing the detection signal of each of the multiple time domains, a determination unit, which integrates and learns a plurality of outlier detection algorithms based on the detection signal value of each time domain and the confirmed feature variables of each time domain, and selects the optimal outlier detection algorithm of each time domain, The detection unit determines the abnormal state of the target object by applying a preset abnormal value detection algorithm to each of the multiple time domains.

2. The sensor outlier detection system using an integrated algorithm taking into account characteristic variables according to claim 1, characterized in that: The monitoring target object is one of a water pipe, an industrial machine tool, a steam turbine running in a power plant, a gas turbine, a wind turbine, a thrust propeller, and an air propeller.

3. The sensor outlier detection system using an integrated algorithm taking into account characteristic variables according to claim 1, characterized in that: The waveform analysis unit converts the detection signal into a frequency domain, and confirms the characteristic variable by analyzing the configuration of frequencies included in the converted detection signal in the frequency domain.

4. The sensor outlier detection system using an integrated algorithm taking into account characteristic variables according to claim 3, characterized in that: The determination unit determines other outlier detection algorithms by considering at least one of the number of frequencies and the intensity of frequencies of characteristic variables contained in the detection signal in the frequency domain.

5. The sensor outlier detection system using an integrated algorithm taking into account characteristic variables according to claim 1, characterized in that: The characteristic variables include: at least one of the vibrations generated by other mechanical equipment around the monitoring target during initial operation, the vibrations generated due to the continuous operation of the other mechanical equipment, the vibrations based on the output using the other mechanical equipment, and the vibrations of vehicles passing by.

6. The sensor outlier detection system using an integrated algorithm taking into account characteristic variables according to claim 1, characterized in that: The detection unit does not perform the determination of the abnormal state in a time domain in which the change of the waveform in the plurality of time domains is continuously lower than a standard value.

7. A sensor abnormal value detection method using an integrated algorithm that considers characteristic variables, the sensor abnormal value detection method detects abnormal values ​​of a sensor of a monitoring device that monitors an abnormal state of a target object, characterized in that: include: The time analysis unit of the monitoring device accumulates and stores the vibration detection signal of the sensor provided at the target object, The time analysis unit analyzes the change of the waveform of the detection signal according to time in units of days, and divides the time period having the signal similarity within the set standard into several steps, The waveform analysis unit of the monitoring device analyzes the detection signal of each time domain in the plurality of time domains to confirm the characteristic variable, The determination unit of the monitoring device integrates and learns multiple outlier detection algorithms based on the detection signal value of each time domain and the confirmed feature variables of each time domain, and determines the optimal outlier detection algorithm for each time domain; and The detection unit of the monitoring device applies a preset abnormal value detection algorithm to each of the multiple time domains to determine the abnormal state of the target object.

8. The sensor abnormal value detection method using an integrated algorithm considering characteristic variables according to claim 7, characterized in that: The monitoring target object is any one of a water pipe, an industrial machine tool, a steam turbine running in a power plant, a gas turbine, a wind turbine, a thrust propeller, and an air propeller.

9. The sensor abnormal value detection method using an integrated algorithm considering characteristic variables according to claim 7, characterized in that: The waveform analysis unit converts the detection signal into a frequency domain, and confirms the characteristic variable by analyzing the configuration of frequencies included in the converted detection signal in the frequency domain.

10. The sensor abnormal value detection method using an integrated algorithm considering characteristic variables according to claim 9, characterized in that: The determination unit determines other outlier detection algorithms by considering at least one of the number of frequencies or the intensity of frequencies of characteristic variables contained in the detection signal in the frequency domain.

11. The sensor abnormal value detection method using an integrated algorithm considering characteristic variables according to claim 9, characterized in that: The characteristic variables include: at least one of the vibration or noise generated by other mechanical equipment around the monitoring target during initial operation, the vibration or noise generated due to the continuous operation of the other mechanical equipment, and the vibration or noise based on the output of the other mechanical equipment.

12. The sensor abnormal value detection method using an integrated algorithm considering characteristic variables according to claim 8, characterized in that: The detection unit does not perform the determination of the abnormal state in a time domain in which the change of the waveform in the plurality of time domains is continuously lower than a standard value.

Images