KR102141391B1 - Failure data management method based on cluster estimation - Google Patents

Abstract

The present invention relates to a technology for recording and managing failure data occurring in facilities. A method for managing failure data for a target facility by a failure recording system collects failure indicator data in a situation where a failure has occurred for each function location to observe whether a failure has occurred with respect to the target facility, to match and store the failure indicator data with the failure history, causes, and types, extracts a plurality of predetermined feature values from the failure indicator data to calculate a degree of cluster for the failure indicator data, determines a level of clarity of the evaluation criteria for each failure in accordance with the calculated degree of cluster, judges whether the determined level of clarity falls within a predetermined threshold range, and induces reclassification of the corresponding phenomenon in accordance with the judgment result.

Description

Failure data management method based on cluster estimation

The present invention relates to a technique for recording failure data occurring in production and operation facilities, and more particularly, to a method for managing failure data to re-classify the failure through cluster evaluation by measuring indicator data indicating the state of the target equipment.

The monitoring and recording of status, and the target facility diagnosis technology, are related to facility maintenance activities such as determining the current maintenance status by accumulating the history of equipment and determining the maintenance time of equipment through data analysis. In addition to being important, facility preservation activities are essential activities that must be carried out for long-term operation, and have been in the spotlight as a core technology area that increases not only the productivity of facilities but also reliability and stability.

For example, many of the first phenomena that occur when a failure occurs from an operating mechanical installation are vibrations. Throughout the literature, it has long been reported that vibration occurs as a major symptom associated with mechanical defect problems such as wear, malfunction, noise and structural damage. The vibration generated by the machine is gradually transferred to other parts during the deterioration process to generate secondary vibrations, or some of the vibrational energy is emitted as noise. Therefore, by periodically measuring and recording vibrations and noises and analyzing them, it is possible to identify the internal condition of the machine and to prevent mechanical defects that can cause fatal damages early by determining abnormalities, and provide important information in establishing maintenance strategies. Can provide.

As described above, in recent years, industrial sites monitor various indicator data indicating the health of mechanical equipment, diagnose the cause of the failure using the measured indicator data, and stop the machine through proper equipment maintenance activities. Efforts are being made to reduce the cost of down-time. The prior art documents presented below suggested a series of inspection procedures for diagnosing failures after classifying and structuring the types of failures occurring in the machine tool.

On the other hand, in classifying the types of faults, it is a general aspect of fault data management that the classification criteria are established by skilled workers who have a lot of experience in the field, and the fault data is measured and recorded according to the established classification. However, in order to obtain quantified failure data in a situation in which a maintenance person performing a failure diagnosis records a failure, a system is used in which a failure is predefined and a failure is selected from a failure category list. Such a system is a system for recording the behavior of a person, and it is highly likely that the accumulated data will not be normal when the selection criteria are different or ambiguous, and in the end, the analysis caused by the data that is not normal may adversely affect the results. have.

Therefore, a system for converting human behavior to data periodically needs technical means to evaluate whether the categories of data are properly defined and improve the normality of data.

Korean Patent Publication No. 2006-0077814, "Diagnostic system and method for machine tools"

The technical problem to be solved by the present invention is to solve the problem that data accumulated in the failure recording system loses its normality due to different or ambiguous selection criteria for each person when conventional technologies select and record the currently occurring failure. In order to overcome the limitation, a situation in which a predefined failure category in the failure recording system is inappropriate or loses the effectiveness of the failure category over time.

In order to solve the above technical problem, a method for managing failure data for a target facility by a failure recording system including at least one processor and data storage means according to an embodiment of the present invention includes: (a) failure Collecting, by the recording system, failure indicator data in a situation where a failure occurs for each functional location to observe whether the equipment has a failure, matching and storing the failure history, the cause and the type together; (b) the failure recording system extracting a plurality of preset feature values from the failure indicator data to calculate a clustering degree for the failure indicator data; (c) determining a level of clarity of an evaluation criterion for each failure according to the clustering in which the failure recording system is calculated; (d) the failure recording system determining whether the determined level of clarity falls within a predefined threshold range; And (e) inducing the failure recording system to reclassify the corresponding failure phenomenon in any one of'maintenance','separation', and'merge' according to the determination result.

In the method of managing the failure data according to an embodiment, the step (a) of collecting the failure indicator data and matching the failure history, the cause, and the type, and storing the (a1) is actually performed among the types of the predefined failure phenomenon. Selecting a failure phenomenon corresponding to the failure; (a2) storing state data indicating performance of a function including at least one of vibration, noise, temperature, flow rate, and pressure of a facility corresponding to the selected failure phenomenon, and matching the selected failure phenomenon; And (a3) storing, in addition to the state data of the facility, information that is the basis for determining the failure, and matching the selected failure phenomenon.

In the method of managing failure data according to an embodiment, the step (b) of calculating a clustering degree of the failure indicator data includes: (b1) separating a predetermined frequency band from the failure indicator data and time domain in the frequency domain. Converting to and performing envelope analysis for extracting the original signal excluding the transmission wave in the time domain, shaping the data through signal processing; further comprising extracting a plurality of feature values from the standardized failure indicator data Can.

In the method of managing failure data according to an embodiment, the step (b) of calculating the clustering degree of the failure indicator data includes (b2) each feature based on clustering and differentiation of classification for a plurality of preset feature values It may further include the step of measuring the sensitivity of, and selecting only the characteristic whose measured sensitivity is greater than or equal to a reference value as a feature value for diagnosing a failure. In addition, the step (b) of calculating the clustering degree of the failure indicator data may further include (b3) performing scaling to adjust the range and variance of all selected feature values to a constant size. have.

In the method of managing failure data according to an embodiment, the step (c) of determining the level of clarity of the evaluation criteria for each failure includes clustering feature values in one failure based on the calculated clustering degree. ) By evaluating clustering differentiation.

Further, the step (d) of determining whether the level of clarity falls within a predefined threshold range includes (d2) the clarity of the failure type and the first predefined threshold range based on the evaluated clustering differentiation. Determining whether a fault phenomenon is necessary by comparing; And (d2) cross-comparison a plurality of suspected faulty items in the list of suspected faults generated according to the newly proposed function location and observation phenomenon, and define in advance the clarity and clarity between the faulty items based on the clustering and differentiation of the corresponding faulty items It may include; determining whether to merge the failure phenomenon by comparing the second threshold range.

Further, the step (e) of inducing the re-classification of the failure phenomenon may include: (e1) separating the corresponding failure item into two based on the evaluated clustering discrimination when it is determined that'separation' of the failure phenomenon is necessary; (e2) merging two corresponding failure items into one based on the evaluated clustering differentiation when it is determined that'merging' of the failure phenomenon is necessary; Or (e3) maintaining a corresponding failure item when it is determined that neither the'separation' nor the'merge' is necessary. Among them, it is possible to induce a failure phenomenon to be reclassified by selectively performing either method.

Meanwhile, hereinafter, a recording medium readable by a computer recording a program for executing the above-described failure data management methods on a computer is provided.

In the embodiments of the present invention, the feature value is unclear by extracting the feature value of the indicator data indicating the feature of the failure and inducing the reclassification of the failure when the feature of the failure violates the normality by measuring the clustering of the extracted feature value distribution. It reduces the diagnostic error caused by one failure and improves the reliability of the analysis results to which the failure data is applied.

1 is a view for explaining the overall process of performing a maintenance according to the diagnosis, record the failure of the failure record system is implemented in the embodiments of the present invention, the failure record.
2 is a block diagram showing a failure recording system for managing failure data of a target facility based on a cluster evaluation of failure data according to an embodiment of the present invention.
3 is a flowchart illustrating a method of managing failure data of a target facility based on a cluster evaluation of failure data according to an embodiment of the present invention.
4 is a flowchart illustrating in more detail the process of collecting and storing the failure indicator data in the method of managing the failure data of FIG. 3 according to an embodiment of the present invention.
5 is a diagram illustrating a situation in which a category of a failure phenomenon is selected.
6 is a flowchart illustrating in more detail the process of calculating the clustering degree in the method of managing fault data of FIG. 3 according to an embodiment of the present invention.
7 to 9 are diagrams for explaining a signal processing process for collected failure indicator data.
10 is a flowchart more specifically illustrating a process of reclassifying a failure phenomenon in consideration of a level of clarity in the method of managing the failure data of FIG. 3 according to an embodiment of the present invention.
11 is a view for explaining a process for determining whether separation of a failure phenomenon is necessary within one failure type.
12 is a diagram for explaining a process of determining whether merging of failures is necessary between two types of failures.

Prior to describing the embodiments of the present invention, the technical means adopted by the embodiments of the present invention are introduced in general, and then specific components are sequentially described.

Equipment failure data is useful for equipment maintenance and asset management. For example, a representative value may be calculated from a distribution of occurrence times of failure events recorded in the past to schedule occurrence times of failures. In addition, failure data can be used as labeling data of supervised learning to derive a machine learning model that can determine the integrity of facilities and functional integrity and predict the life of the equipment.

In order to acquire such failure data, the'maintenant (one who performs the fault diagnosis)' diagnoses what kind of failure occurred by comparing the characteristics of the predefined failure and the current situation of the equipment when the equipment failure occurs. And record the results. The procedure for recording the fault diagnosis result can be largely composed of two steps.

First, the'operator' or'observer' of the facility notifies the'maintenant' of the'functional location' and'observation phenomenon' in which the failure has been observed since the point of view of the failure. Second, the'suspect failure' list determined according to the notified'function location' and'observation phenomenon' is removed from the list or the'failure phenomenon' according to the test execution result corresponding to each'suspect failure' It can be confirmed as.

However, as described above, in the situation of recording a fault, when the current fault is selected and recorded from a list of predefined fault categories, data accumulated in the fault recording system is normalized due to different or ambiguous selection criteria for each person. The problem was pointed out. In addition, the question arises as to whether the predefined failure category is correctly defined in the failure recording system, and further, over time, whether the previously defined failure category is still valid. This is because there are cases where the classification of the failure category needs to be changed according to the accumulation of failure data.

Therefore, the embodiments of the present invention described below do not use a fixed type of failure category, but also collects indicator data to check the state of the facility when recording a failure, and collects indicator data for each failure recorded in the past. Evaluate the clustering level of and raise the need for re-classification/re-definition of failures under conditions that do not meet the threshold (critical range). We would like to propose a technical means to find the feature values that show differentiation within the same failure and induce the separation of the failures, or to find the aggregates showing different clusters between different failures and induce the merging of failures.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. However, in the following description and attached drawings, detailed descriptions of well-known functions or configurations that may obscure the subject matter of the present invention are omitted. In addition, throughout the specification,'comprising' a component means that other components may be further included rather than excluding other components, unless otherwise stated.

The terms used in the present invention are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms “include” or “have” are intended to indicate that a feature, number, step, action, component, part, or combination thereof is described, and that one or more other features or numbers are present. It should be understood that it does not preclude the existence or addition possibility of steps, actions, components, parts or combinations thereof.

Unless specifically defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains. Terms, such as those defined in a commonly used dictionary, should be interpreted as meanings consistent with meanings in the context of related technologies, and should not be interpreted as ideal or excessively formal meanings unless explicitly defined in the present application. . In the following, the proposed failure diagnosis method was verified by applying it to the LNG facility from the viewpoint of implementation and simulation, but this is for illustration only, and it is natural that the target facility is not limited thereto.

In an environment in which a fault diagnosis system according to embodiments of the present invention is implemented, first, a fault of a facility is defined and a fault is recorded when a fault occurs. The occurrence time of the failure can be scheduled by calculating a representative value from the distribution diagram of the past failure occurrence time. In addition, a machine learning model capable of diagnosing the condition of a facility from the indicator data using the indicator data and the failure information may be calculated using supervised learning. For the indicator data, for example, data that can indicate the state of the facility, such as vibration, temperature, pressure, and operation time of the facility, may be used.

When recording a failure, the indicator data to check the condition of the equipment is collected together, and the clustering of the indicator data is evaluated for each failure recorded in the past, and the need for reclassification/re-definition of the failure under conditions that do not reach the standard value (critical range) By inspecting, we induce the separation of failures for feature values that show differentiation within one failure, or find a set that shows clustering between two failures and induce mergers of failures.

In addition, when recording a fault, a list of suspected faults according to the observed phenomenon is shown, and the feature values are extracted from the index data of past recorded faults through the process of removing the suspected faults one by one. Rearrange the fault list. For this feature value, for example, at least one of the average, variance, moment, and entropy of the data can be used. In addition, the clustering evaluation can be calculated as the average of the relative distances of feature values determined to be the same failure.

A failure in which the distribution of feature values is concentrated means that there is a clear characteristic when determining the failure. Therefore, since the verification is performed in the order of specific faults in the rearranged suspected fault list, the accuracy of diagnosis and the time required can be shortened.

1 is a view for explaining the overall process of performing a maintenance according to the diagnosis, record the failure of the failure record system is implemented in the embodiments of the present invention, the failure record. The specific meanings of the terms indicated in FIG. 1 can be confirmed more accurately through Table 1 below.

Referring to FIG. 1 and Table 1 together, the function location and the type of the fault are first matched and input together with the indicator data on the fault (110). In the input process, the clustering of indicator data is evaluated for each failure recorded in the past, and if it does not reach the reference value (threshold range), the failure is reclassified. Then, a list of suspected faults is read through a verification process, and each faulty item is examined to exclude or confirm (120). Now, through the inspection (inspection) process, the check list corresponding to the previously determined suspect failure list is checked to identify the damage mechanism (130). That is, the damage mechanism is identified for the identified failure in the verification process, and the identified failure may be composed of a sequence pair of'functional location' and'fault type'. Fault diagnosis is completed through the above process, and the embodiments of the present invention focus on the process 110 of collecting and recording indicator data related to a fault among the above-described processes.

On the other hand, if a broken damage mechanism is identified, maintenance is performed according to predefined measures for the mechanism (140). Finally, the final quality is verified through an operation check on whether maintenance is properly performed (150).

2 is a block diagram showing a failure recording system 200 for managing failure data of a target facility based on a cluster evaluation of failure data according to an embodiment of the present invention.

First, the failure indicator data in a situation where a failure occurs is collected by a function location to be observed for failure from the target facility, and matched together with the failure history, cause, and type, and received through the input unit 210. This indicator data is a leading indicator that can identify the cause of the failure, and corresponds to raw data that is the basis of the failure diagnosis. The collected indicator data may be stored in the storage unit 230 which is a data storage means of the failure recording system 200.

The input indicator data is preferably managed as a digital signal from the viewpoint of data processing, and can be processed under the control of software consisting of a set of instructions for a series of fault recording processes through at least one processor 250. have. To this end, the processor 250 extracts a plurality of preset feature values from the failure indicator data, calculates a clustering degree for the failure indicator data, and clarifies evaluation criteria for each failure according to the calculated clustering degree. A degree level is determined, and it is determined whether the determined level of clarity falls within a predefined threshold range, and a corresponding failure phenomenon is determined according to the determination result, one of'maintain','separate', and'merge' Induce them to reclassify in this way. Finally, the failure recording system 200 may generate a result in which the failure phenomenon is reclassified. The reclassified failure phenomenon may be used as labeling data of supervised learning, and furthermore, it may be used for predicting the failure or predicting the remaining life.

FIG. 3 is a flowchart illustrating a method of managing failure data of a target facility based on a cluster evaluation of failure data according to an embodiment of the present invention, and reconstructs the configuration of FIG. 2 described above in a time series sequence. Accordingly, the method for managing the fault data of FIG. 3 may be implemented with software including instructions for performing a series of processes described below, and a fault recording system including at least one processor and data storage means It can be mounted and driven.

In step S310, the failure recording system collects failure indicator data in a situation where a failure occurs for each functional location to observe whether or not there is a failure with respect to a target facility, and stores it by matching with the failure history, cause, and type. That is, the identifier and the failure index data are input to the component that is assigned a unique number for distinguishing the functional location of the target facility and used as a basis for analysis.

In step S320, the failure recording system extracts a plurality of preset feature values from the failure indicator data to calculate a clustering degree for the failure indicator data. At this time, the feature value includes at least one of the average, variance, moment, and entropy of the data, and the clustering is preferably calculated as the average of the relative distances of the feature values determined to be the same failure. The distribution of these feature values is concentrated, and it is possible to determine whether the currently set failure classification is appropriate for one failure through the phenomenon.

In step S330, the failure recording system determines the level of clarity of the evaluation criteria for each failure according to the calculated clustering degree. Clustering indicates how dense the distribution of failure data is, and failures with dense distribution of feature values mean that there are clear features when determining failures, so whether each failure is properly classified through clustering evaluation The clarity of the evaluation criteria for can be determined. To this end, in this process, clustering discrimination can be evaluated by clustering feature values in one failure based on the clustering calculated through step S320.

More specifically, the process of determining the clarity of the evaluation criteria for each failure numerically according to the clustering of the indicator data using the extracted feature values is as follows. First, clusters generated by clustering feature value distributions within one'fault type' are designated as classes. Then, it is inversely proportional to the average of the density inside all classes, and is computed in proportion to the distance or metric average of the center values of the different classes to calculate the clustering for the'fault type' numerically. By doing so, the level of clarity of the evaluation criteria can be determined. On the other hand, by cross-comparing multiple failures in the list, the ratio of the average of the density of two corresponding'fault types' and the distance or metric to each other is calculated for the two'fault types'. Differentiation can be calculated numerically.

In step S340, the failure recording system determines whether the determined level of clarity falls within a predefined threshold range. At this time, the threshold range is a criterion for determining whether the current failure classification is appropriate, for example, it is determined whether separation of a failure phenomenon is necessary within one failure type, or a failure between two failure types It is used as a basis for determining whether merging of phenomena is necessary.

Then, in step S350, the failure recording system induces to re-classify the corresponding failure phenomenon in any one of'maintenance','separation', and'merge' according to the determination result through step S340.

When the clustering for the'fault type' calculated through the step S330 is compared with the threshold range for the clustering in the step S340, if the deviation exceeds the allowable range, the user is given a'separation', and the names of the two'fault types' are separated. You can set a new one. On the other hand, if the difference between the two'fault types' calculated through step S330 is compared to the threshold range for the difference in step S340, the user suggests the'combination' of the two'fault types' when it falls outside the allowable range, The names of the two'fault types' combined can be set.

In performing these steps S340 and S350, a series of computational processes for determining are involved, which will be described in more detail with reference to FIGS. 10 to 12 later.

On the other hand, if the failure data is accumulated over a period of time, the previously re-classified failure phenomenon may not be valid again. In this case, it is preferable to induce the failure to be reclassified according to the changed failure data by repeatedly performing the above steps S310 to S350.

4 is a flowchart illustrating in more detail the process of collecting and storing the failure indicator data (step S310) in the method of managing the failure data of FIG. 3 according to an embodiment of the present invention.

First, in step S311, a failure phenomenon corresponding to an actual failure is selected from the types of predefined failure phenomena. Accordingly, tangible information corresponding to various types of indicator data related to the failure can be provided through the failure recording system. Referring to FIG. 5 exemplifying a situation in which a category of a fault phenomenon is selected, a screen displaying various types that can be assigned to a fault phenomenon currently defined in the fault recording system can be seen. Of course, when the first system is operated, a type of a predefined failure phenomenon may be preset by a user, and then a reclassification of a category occurs as a record of failures is accumulated in the system.

In step S312, state data representing performance of a function including at least one of vibration, noise, temperature, flow rate, and pressure of a facility corresponding to the failure phenomenon selected through step S311 is stored, and matching with the selected failure phenomenon do. That is, the indicator data for the failure phenomenon is matched and stored.

In step S313, in addition to the state data of the facility, information that is the basis for determining the failure is stored, and the selected failure phenomenon is matched. That is, additional data related to the failure determination are linked and stored.

6 is a flowchart more specifically illustrating a process (step S320) of calculating a clustering degree in the method of managing fault data of FIG. 3 according to an embodiment of the present invention.

In step S321, the predetermined frequency band is separated for the failure indicator data, the time domain is converted to the frequency domain, and envelope analysis is performed to extract the original signal excluding the transmission wave from the time domain, thereby performing data through signal processing. By normalizing, a plurality of feature values can be extracted from the standardized failure indicator data. A series of signal processing processes performed in step S321 will be described in more detail with reference to FIGS. 7 to 9 illustrating the signal processing process for the collected fault indicator data.

The collection and extraction of data containing important information in determining the condition of the target facility can be obtained through an appropriate signal processing process for the measured signal. For example, the vibration signal generated by the machine is composed of many frequency components related to each component of the machine, and thus has various and complex characteristics. The appropriate signal processing technique for the vibration signal is largely divided into a time domain and a frequency domain, and signal processing methods for each domain are applied differently.

First, a Discrete Wavelet Transform (DWT) can be utilized to separate a specific frequency band from a time domain signal using a low pass filter and a high pass filter. At this time, the result of passing through the low-pass filter is approximation, and the result of passing through the high-pass filter is detail coefficients. As the result of passing through each filter is half the frequency, half of the sample can be removed.

Referring to FIG. 7(A), which shows a process of separating a frequency band with respect to the index data, this decomposition process can be repeated in a stepwise manner to increase frequency resolution and split a required frequency band. It can be represented in the form of a binary tree, and each node is represented by a time-frequency. FIG. 7B shows frequency bands at each level of FIG. 7A.

Second, Fast Fourier Transform (FFT) can be used to transform the time domain to the frequency domain. 8 is a diagram illustrating a time domain waveform (A) and a frequency domain waveform (B) of index data, respectively.

Third, an analytic signal using a Hilbert-Huang Transform (HHT) can be used as an envelope analysis method for extracting an original signal excluding a transmission wave in the time domain. Referring to FIG. 9, an envelop of the signal can be obtained by generating an imaginary pair having the same magnitude as the original signal but having an orthogonal phase with the Hilbert transform.

Returning to FIG. 6 again, in step S322, the sensitivity of each feature is measured based on the clustering and differentiation of the classification for a plurality of preset feature values (feature vectors), and only the feature whose sensitivity is higher than the reference value is used for fault diagnosis. It can be selected as a feature value (feature vector). In addition, the number of feature values (feature vectors) may be determined according to the result of the classification methodology used while varying the number of the selected feature values (feature vectors) for each rank.

First, in the time domain, in order to extract characteristics such as energy, distribution, and amplitude of the measured signal, as shown in Tables 2 and 3 below, feature parameter values related to failure diagnosis of the target facility may be used. Table 2 describes time domain parameters and their meanings, and Table 3 describes frequency domain parameters and their meanings.

In the case of the feature value (feature vector) acquired through the signal processing process, it is generally high-dimensional, and when it is used as an input of a classifier for machine learning, it may cause a decrease in classification performance. Therefore, techniques for extracting only effective feature values (feature vectors) are essential for reliable fault diagnosis.

In this example, the strategy of measuring the sensitivity of each feature on the basis of clustering and differentiation of class through Distance Evaluation Technique (DET) and selecting only the features that are effective in diagnosing failures based on this. According to the DET, the average distance between samples in a class is calculated for each feature of a failure in a matrix consisting of a plurality of classes, failure characteristics, and a certain number of samples for each class, and the degree of clustering of all classes for each feature is calculated. Calculate. Then, the degree of discrimination between classes for the feature can be derived to obtain the sensitivity values for the entire class of the feature. Now, it is possible to exclude feature values (feature vectors) that are inefficient for diagnosis by selecting only features whose measured sensitivity is greater than or equal to a reference value as a feature value (feature vector) for diagnosis of failure.

Returning to FIG. 6 again, in step S323, scaling may be performed to adjust the range and variance of all selected feature values to a constant size. Since the range and variance of each feature value (feature vector) derived above are different from each other, when performing classification (for example, using SVM) in a multidimensional space, despite the feature having a relatively small cluster value, It may happen that the classification is based on a feature with a large range. Therefore, it is preferable that a scaling process is performed to uniformly adjust the range and variance of all selected features before performing classification (eg, application of SVM). For example, normalization may be used as the scaling method.

10 is a flowchart more specifically illustrating a process of reclassifying a failure phenomenon (step S350) in consideration of a level of clarity (step S340) in the method of managing the failure data of FIG. 3 according to an embodiment of the present invention.

Earlier, clustering discrimination was evaluated by clustering feature values in one failure based on the clustering calculated through step S330. Therefore, in step S340, a process of determining whether separation or merging is necessary in case of one failure or two failures is presented.

In step S341, it is determined whether separation of the failure phenomenon is necessary by comparing the clarity of the failure type and a predefined first threshold range based on the clustering difference evaluated in step S330. More specifically, in step S330, clusters generated by clustering feature value distributions within one'fault type' are designated as classes, inversely proportional to the average of the density inside all classes, and the distance from the center values of different classes It was described that the clustering for the corresponding'fault type' can be calculated numerically with a calculation formula proportional to the average. Now, in step S341, the calculated clustering value is compared with the first threshold range. Here, the first threshold range may be set as an arithmetic value or a section that can be compared with a cluster value as a value or section that is a criterion for determining whether to separate the corresponding fault according to clustering differentiation.

In step S342, a plurality of suspected faulty items existing in the list of suspected faults generated according to the newly proposed function location and observation phenomenon are cross-comparison to define and clarify and clarify between faulty items based on the clustering and differentiation of the corresponding two faulty items. By comparing the second threshold range, it is possible to determine whether or not a fault phenomenon needs to be merged. More specifically, it was described in step S330 that the difference between the mean of the density of two'fault types' and the distance between each other can be calculated numerically for the two'fault types'. Now, in step S342, the calculated differential value is compared with the second threshold range. As briefly introduced through FIG. 1, embodiments of the present invention may provide a list of suspected failures according to a given function location and observation phenomenon, and the list of suspected failures is a plurality of suspected failures that are presumed to have occurred. Includes items. That is, when determining whether classification between currently set faults is appropriate, two fault types that are likely to be included together in a list of suspect faults (high correlation) are selected, rather than randomly selecting two fault types. Thus, a cross comparison is performed between the two. At this time, whether to merge or not is determined by comparing with the second threshold range how much the two are related and how clear the two are based on the clustering and the difference between the failure items.

Now, in step S350, separation/merge/maintenance of the failure site is determined according to the determination result (step S340). Step S350 is induced to be performed according to any one of the following three calculation processes.

When it is determined in step S341 that the'separation' of the failure phenomenon is necessary, the process proceeds to step S351 to separate the corresponding failure item into two based on the evaluated clustering discrimination. If it is determined in step S342 that the'merging' of the failure phenomenon is necessary, the process proceeds to step S352 and merges two corresponding failure items into one based on the evaluated clustering differentiation. If it is determined that neither of the'separation' and the'merging' is necessary, the failure item is maintained in step S353.

11 is a view for explaining a process for determining whether separation of a failure phenomenon is necessary within one failure type. Referring to Figure 11, (A) → (B) → (C), deriving the clustering differentiation in the order of (D), and illustrated a process for determining whether it is necessary to separate for one failure phenomenon. FIG. 11C shows a process of calculating an average distance between samples in a class for each feature of a failure, and calculating the degree of clustering of the entire class for each feature. In the case of the illustrated class 4, it can be seen that the distance between samples is relatively distant and the density is low. 11(D) shows a process of obtaining a sensitivity value for all classes of a corresponding characteristic by deriving a degree of distinction between classes for the characteristic. In the case of the illustrated class 4, it can be seen that the distance from other classes is relatively large.

12 is a diagram for explaining a process of determining whether merging of failures is necessary between two types of failures. Referring to FIG. 12, the same failure phenomenon is described on the horizontal axis and the vertical axis, and the results of cross-comparing these are visualized. Through this, it is possible to determine whether it is necessary to merge the faults based on the clustering and differentiation of the two faults. For example, looking at the result of cross-comparison of failures of'burnout/cutting' and'crack/breaking', it can be seen that the clusters of the two are not clearly separated and do not have differentiation from each other. It can be judged that it is necessary to merge the faults. On the other hand, looking at the result of cross-comparison of the failures of'foreign material/contamination' and'burnout/cutting', it can be seen that the clusters of the two are clearly separated and differentiated from each other. It is more desirable not to maintain each type.

The embodiments of the present invention described above, by extracting the feature value of the indicator data indicating the characteristic of the failure and measuring the clustering of the extracted feature value distribution as a measure, induce the re-classification of the failure when the feature of the failure violates normality, It reduces the diagnostic error that will occur due to an unclear failure and improves the reliability of the analysis results to which the failure data is applied.

On the other hand, the present invention can be implemented in the computer-readable code on the computer-readable recording medium embodiments. The computer-readable recording medium includes all kinds of recording devices in which data readable by a computer system is stored.

Examples of the computer-readable recording medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. In addition, the computer-readable recording medium can be distributed over network coupled computer systems so that the computer-readable code is stored and executed in a distributed fashion. And functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the technical field to which the present invention pertains.

In the above, the present invention has been mainly focused on the various embodiments. Those skilled in the art to which the present invention pertains will appreciate that the present invention may be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered in terms of explanation, not limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent range should be interpreted as being included in the present invention.

200: fault recording system
210: input
230: storage
250: processor

Claims (10)

A method of managing a failure data for a target facility by a failure recording system having at least one processor (processor) and data storage means,
(a) collecting a failure indicator data in a situation where a failure occurs for each functional location where a failure recording system wants to observe a failure with respect to a target facility, and storing it by matching with a failure history, cause, and type;
(b) the failure recording system extracting a plurality of preset feature values from the failure indicator data to calculate a clustering degree for the failure indicator data;
(c) Clusters of characteristic values from one failure based on the cluster diagram calculated by the failure recording system are inversely proportional to the average of the density inside all clusters occurring within the failure type, and different clusters Determining a level of clarity of an evaluation criterion for each failure by calculating a clustering value to be proportional to a distance average of the center values of the;
(d) the failure recording system determining whether the determined level of clarity falls within a predefined threshold range; And
(e) inducing the failure recording system to reclassify the corresponding failure phenomenon in one of'maintenance','separation', and'merge' according to the determination result;
Step (d) is,
(d2) determining whether a failure phenomenon is necessary to be separated by comparing the clarity of the failure type and a first predefined threshold range based on the calculated cluster value; And
(d2) Cross-comparison a plurality of suspected faulty items in the list of suspected faults generated according to the newly proposed function location and observation phenomenon, and define the clarity and predefinedness between faulty items based on the clustering and differentiation of the two faulty items. And determining whether to merge the faults by comparing the second threshold range. According to claim 1,
Step (a) is,
(a1) selecting a failure phenomenon corresponding to an actual failure from among predefined types of failure phenomenon;
(a2) storing state data indicating performance of a function including at least one of vibration, noise, temperature, flow rate, and pressure of a facility corresponding to the selected failure phenomenon, and matching the selected failure phenomenon; And
(a3) storing information that is the basis for determining the failure in addition to the state data of the facility, and matching with the selected failure phenomenon. According to claim 1,
Step (b) is,
(b1) By performing a envelope analysis for separating a predetermined frequency band, converting a time domain into a frequency domain, and extracting an original signal excluding a transmission wave from the time domain, the data is processed through signal processing. The step of normalizing; extracting a plurality of feature values from the standardized failure indicator data by further including, the failure data management method. According to claim 1,
Step (b) is,
(b2) measuring the sensitivity of each feature based on the clustering and differentiation of the classification for a plurality of preset feature values, and selecting only features having a measured sensitivity higher than or equal to the reference value as a feature value for failure diagnosis; , Failure data management method. The method of claim 4,
Step (b2) is,
Calculate the average distance between samples in a class for each failure characteristic in a matrix composed of a number of classes, failure characteristics, and a certain number of samples for each class according to DET (Distance Evaluation Technique) The method of managing failure data, which calculates the degree of clustering of all classes, derives the degree of distinction between classes for the characteristic, and obtains a sensitivity value for the entire class of the characteristic. The method of claim 4,
Step (b) is,
(b3) performing scaling to adjust the range and variance of all selected feature values to a certain size; further comprising a method of managing fault data. delete According to claim 1,
Step (d2) is,
Cross-comparison the two selected fault types by considering the correlation within the suspect fault list rearranged in the order of clustering of feature values extracted from the failure index data, including a plurality of suspect fault items that are estimated to have occurred. How to manage fault data. According to claim 1,
Step (e) is,
(e1) when it is determined that'separation' of the failure phenomenon is necessary, separating the corresponding failure item into two based on the evaluated clustering differentiation;
(e2) merging two corresponding failure items into one based on the evaluated clustering differentiation when it is determined that'merging' of the failure phenomenon is necessary; or
(e3) maintaining the corresponding failure item when it is determined that neither the'separation' nor the'merge' is necessary;
Among them, a method of managing fault data, which induces to reclassify the fault phenomenon by selectively performing one of the methods. According to claim 1,
When the failure data accumulates after a predetermined time has elapsed, the steps (a) to (e) are repeated to induce the failure to be reclassified according to the changed failure data, and the failure data management method.

Images