JP5895457B2 - Anomaly monitoring system and anomaly monitoring method - Google Patents

Description

  The present invention relates to an abnormality monitoring system and an abnormality monitoring method for detecting an abnormality of a plant facility based on time series data obtained from the plant facility.

  In the state monitoring of a plant facility, an abnormality in the plant facility is generally detected by setting appropriate upper and lower limits for signal data obtained from the plant facility. However, when some specific devices such as valves are to be monitored, state monitoring is performed by extracting feature amounts corresponding to the characteristics of the devices.

  However, in a plant configured by combining a plurality of element devices (equipment) to execute one purpose, for example, a steel product manufacturing plant, the above-described state monitoring may be insufficient. For example, in such a plant, each facility often repeats the same operation. For this reason, it is necessary to extract the feature amounts such as the rate of change of the signal data, the maximum value, the minimum value, and the time required for settling from each operation pattern, and monitor the operation pattern by managing the extracted feature amounts. In addition, for equipment with a fixed operation pattern, instead of extracting features, the operation pattern may be monitored by monitoring the operation pattern by setting the upper and lower limits according to the manual operation by the user, and an abnormality may be determined. (For example, refer to Patent Document 1).

  On the other hand, a state monitoring technique using multivariate analysis is also known as an approach from another viewpoint. For example, it is known that a principal component analysis is performed on a plurality of process quantities collected from monitored plant equipment, and the process quantities are converted into a small number of feature quantities representing the main changes (for example, (See Patent Document 2).

Japanese Patent Laid-Open No. 10-6028 JP 2001-75642 A

  However, in the method of setting the upper and lower limits for the signal data obtained from the monitoring target as described above and the method of setting the upper and lower limits for the operation pattern as in Patent Document 1, the plant is configured by a plurality of facilities. In such a case, the upper and lower limits must be individually set for each of the plurality of facilities, and there is a problem that manpower and cost increase. On the other hand, in the method of Patent Document 2, since information in the time direction cannot be utilized, there is a case where the accuracy of abnormality detection is lowered in the state monitoring of equipment having an operation pattern.

  The present invention has been made in view of the above, and can be applied universally without being conscious of the type of monitoring target and the like, and an abnormality monitoring system and abnormality monitoring method capable of detecting the abnormality with high accuracy The purpose is to provide.

  In order to solve the above-described problems and achieve the object, the abnormality monitoring system according to the present invention is based on time series data obtained from constant operation equipment that repeatedly performs a predetermined operation during normal operation in a manufacturing process. An abnormality monitoring system for detecting an abnormality of equipment, wherein L is obtained from each of time-series data of a section in which M types (M ≧ 1) of predetermined operations obtained from the constant operation equipment during normal operation are performed in advance. Sampled, acquired as M × L type variables, points in the M × L dimensional space determined by the M × L type variables as normal patterns, and the phase between each variable of a plurality of normal patterns in the space A normal pattern extracting means for calculating the number of relations, and if the maximum value of the correlation coefficient is less than a predetermined threshold value, the mean and variance of each variable are calculated using the constant motion equipment as a first model, and the correlation Maximum number In the case of a predetermined threshold value or more, a monitoring index creating means for calculating principal component conversion coefficients by performing principal component analysis on each variable using the constant operation facility as a second model, and from the constant operation facility during operation The M × L type variables are acquired from the obtained time series data of the section in which the M types of predetermined operations are performed, and points in the M × L dimensional space determined by the M × L type variables are operated. The operation pattern extraction means as a time pattern, and the Mahalanobis distance of the operation pattern based on the average and variance of the normal patterns acquired by the monitoring index creation means when the constant operation facility is the first model Is calculated as a deviation degree, and when the constant operation facility is the second model, from the main component of the operating pattern based on the conversion factor of the main component of the normal pattern acquired by the monitoring index creating means An operating deviation calculating means for calculating a deviation as a deviation; and a determining means for determining an abnormality of the fixed operation facility based on the deviation calculated by the operating deviation calculating means. And

  In the abnormality monitoring system according to the present invention, in the above invention, the determination means determines whether the constant operation facility needs to be repaired based on the number of times that the constant operation facility is determined to be abnormal in a predetermined period. It is characterized by determining.

  In the abnormality monitoring system according to the present invention as set forth in the invention described above, the monitoring index creating means calculates a contribution ratio of principal components of the normal pattern for the constant motion equipment of the second model. The determination means calculates the deviation based on outliers other than the main principal component of the main components of the operating pattern, and the determination means determines that the deviation is equal to or greater than a predetermined value. The operation equipment is determined to be abnormal.

  The abnormality monitoring method according to the present invention is an abnormality monitoring method for detecting an abnormality in the constant operation equipment based on time series data obtained from a constant operation equipment that repeatedly performs a predetermined operation during a normal operation in a manufacturing process. Then, L samples are obtained from each of the time-series data of the section in which the predetermined operation of M types (M ≧ 1) obtained from the constant operation equipment at the time of normal operation is performed, and used as M × L type variables. A normal pattern extracting step of obtaining a point in the M × L dimensional space determined by the M × L type variable as a normal pattern, and calculating a correlation coefficient between each variable of the plurality of normal patterns in the space; When the maximum value of the correlation coefficient is less than a predetermined threshold, the average and variance of each variable are calculated using the constant operation facility as the first model, and the maximum value of the correlation coefficient is equal to or greater than the predetermined threshold No. A monitoring index creation step of calculating principal component conversion coefficients by performing principal component analysis on each variable as a model, and a section in which the M kinds of predetermined operations obtained from the constant operation equipment are performed during operation The M × L type variable is acquired from each of the series data, and the operation time pattern extraction step using the point in the M × L dimensional space determined by the M × L type variable as the operation time pattern, and the constant operation facility Is the first model, the Mahalanobis distance of the operating pattern is calculated as a deviation based on the average and variance of the normal pattern acquired in the monitoring index creating step, and the constant operation facility is the second model. In this case, the deviation from the main component of the operating pattern is calculated as the deviation based on the conversion factor of the main component of the normal pattern acquired in the monitoring index creation step. And extent, characterized in that it comprises a and a determination step of determining an abnormality of the constant operation equipment based on the deviation degree calculated by the operating time deviation calculation step.

  In the present invention, L is sampled and acquired as M × L variables from each of the time-series data of the section in which one or more M types of predetermined operations obtained from the monitoring target at the time of normal operation are performed in advance. It was decided to. Then, a point in the M × L dimensional space determined by the M × L type variable is defined as one normal pattern, and the degree of dispersion of the plurality of normal patterns is calculated. In addition, M × L types of variables are obtained by sampling L times from each of the M types of time-series data obtained from the monitoring target during operation, and within an M × L dimensional space determined by the M × L types of variables. The deviation from the normal pattern was calculated on the basis of the degree of dispersion of the normal pattern with the point of operation as the operating pattern. Then, the abnormality of the monitoring target is determined based on the calculated deviation degree of the operation pattern. According to this, it is possible to provide an abnormality monitoring system and an abnormality monitoring method that can be applied universally without being aware of the type of monitoring target. In addition, it is possible to realize highly accurate abnormality detection in consideration of the pattern of time series data.

FIG. 1 is a block diagram showing an example of the overall configuration of the abnormality monitoring system of the present embodiment. FIG. 2 is a diagram illustrating a typical change curve of two time-series data obtained from the target facility. FIG. 3 is a diagram showing an example of distribution of one time-series data pattern in the L-dimensional space. FIG. 4 is a diagram illustrating a distribution example in the 2L-dimensional space of the operation pattern of the target facility in which two time-series data patterns are combined. FIG. 5 is a flowchart illustrating a processing procedure of the monitoring index creation process. FIG. 6 is a flowchart illustrating a processing procedure of the abnormality monitoring process. FIG. 7 is a diagram illustrating an application example of the embodiment. FIG. 8 is another diagram illustrating an application example of the embodiment. FIG. 9 is another diagram illustrating an application example of the embodiment. FIG. 10 is another diagram illustrating an application example of the embodiment. FIG. 11 is another diagram illustrating an application example of the embodiment. FIG. 12 is another diagram illustrating an application example of the embodiment.

  DESCRIPTION OF EMBODIMENTS Hereinafter, an embodiment for implementing an abnormality monitoring system and an abnormality monitoring method of the present invention will be described with reference to the drawings. Note that the present invention is not limited to the embodiments. Moreover, in description of drawing, the same code | symbol is attached | subjected and shown to the same part.

(Embodiment)
FIG. 1 is a block diagram showing an example of the overall configuration of the abnormality monitoring system 1 of the present embodiment. The abnormality monitoring system 1 according to the present embodiment targets a facility that repeats a certain operation determined in the manufacturing process as a monitoring target, and monitors the status of the monitoring target facility (hereinafter referred to as “target facility”) to detect an abnormality. Is detected. As shown in FIG. 1, the abnormality monitoring system 1 includes an offline index creation system 2, an online diagnostic system 3, an operation DB 4, and a monitoring index DB 5 that are connected to each other so as to be able to transmit and receive data. ing. The operation DB 4 and the monitoring index DB 5 may be stored in the storage units 23 and 33 included in the offline index creation system 2 or the online diagnostic system 3.

  The offline index creation system 2 is realized using a general-purpose computer such as a workstation or a personal computer, and includes an input unit 21, a display unit 22, a storage unit 23, and a control unit 25 that controls each unit.

  The input unit 21 is realized by various input devices such as a keyboard, a mouse, a touch panel, and various switches, and outputs an input signal corresponding to an operation input to the control unit 25. The display unit 22 is realized by a display device such as an LCD, an EL display, or a CRT display, and displays various screens based on a display signal input from the control unit 25.

  The storage unit 23 is realized by various IC memories such as ROM and RAM such as flash memory that can be updated and stored, a hard disk built in or connected by a data communication terminal, an information storage medium such as a CD-ROM, and a reading device thereof. It is. In this storage unit 23, a program for operating the offline index creation system 2 and realizing various functions provided in the offline index creation system 2, data used during the execution of this program, and the like are stored in advance. Or, it is temporarily saved for each processing.

  The control unit 25 is realized by a CPU or the like, based on an input signal input from the input unit 21, a program or data stored in the storage unit 23, and instructions to each unit constituting the offline index creation system 2 Data transfer or the like is performed to control the operation of the offline index creation system 2. The control unit 25 includes a monitoring index creation processing unit 27.

  The monitoring index creation processing unit 27 is a functional unit that performs processing (monitoring index creation processing) for creating a monitoring index used for the status monitoring of the target equipment performed by the online diagnostic system 3 and is obtained from the target equipment during past operations. The normal operating state of the target equipment is indexed using the time series data. The monitoring index creation processing unit 27 includes a partial time series cut-out processing unit 271, a normalization processing unit 273, and a statistical monitoring indexing processing unit 275 as a normal pattern extraction unit and a monitoring index creation unit.

  The online diagnostic system 3 is realized by using a general-purpose computer such as a workstation or a personal computer, as in the offline index creation system 2, and controls the input unit 31, the display unit 32, the storage unit 33, and each unit. Part 35.

  The input unit 31 is realized by various input devices such as a keyboard, a mouse, a touch panel, and various switches, for example, and outputs an input signal corresponding to an operation input to the control unit 35. The display unit 32 is realized by a display device such as an LCD, an EL display, or a CRT display, and displays various screens based on a display signal input from the control unit 35.

  The storage unit 33 is realized by various IC memories such as ROM and RAM such as flash memory that can be updated and stored, a hard disk connected by a built-in or data communication terminal, an information storage medium such as a CD-ROM, and a reading device thereof. It is. In the storage unit 33, a program for operating the online diagnostic system 3 and realizing various functions provided in the online diagnostic system 3, data used during the execution of the program, and the like are stored in advance, or It is temporarily saved for each processing.

  The control unit 35 is realized by a CPU or the like, and based on input signals input from the input unit 31 and programs and data stored in the storage unit 33, instructions and data to each unit constituting the online diagnosis system 3 Is transferred to control the operation of the online diagnostic system 3. The control unit 35 includes an abnormality monitoring processing unit 37.

  The abnormality monitoring processing unit 37 is a functional unit that monitors the state of the target equipment online (in real time) and performs processing for detecting the abnormality of the target equipment (abnormality monitoring processing), and uses time series data obtained from the target equipment. The condition of the target equipment is monitored and an abnormality is detected. The abnormality monitoring processing unit 37 includes a partial time series cut-out processing unit 371, a normalization processing unit 373, a statistical monitoring indexing processing unit 375 as an operation time pattern extraction unit and an operation time deviation degree calculation unit, and a determination And a determination processing unit 377 as means.

  The operation DB 4 stores time-series data acquired from the target equipment during past operations. Also, the monitoring index DB 5 stores a monitoring index in which the monitoring index creation processing unit 27 indexes the normal operation state of the target facility in the offline index creation system 2.

  Next, the principle of state monitoring performed by the abnormality monitoring system 1 will be described. FIG. 2 is a diagram for explaining two variables (hereinafter referred to as “variable A” and “variable B”) determined by two time-series data obtained from the target equipment. In the present embodiment, equipment in a steel product manufacturing plant, specifically, equipment for transporting and positioning a steel product to a predetermined transport position in order to perform processing in the manufacturing process is the target equipment. . This target facility repeatedly performs an operation of sequentially transporting steel products to be transported to a transport position according to the type and the like, and FIG. 2A is obtained by a single operation of the target facility to be repeated. FIG. 2 (b) shows a typical change curve of one time-series data (time-series data defining variable A), and FIG. 2 (b) shows the other time-series data (variable B is obtained by the same operation of the target equipment). A typical change curve of time series data) is shown. More specifically, the time-series data in FIG. 2A is time-series data of, for example, the position of a steel product (hereinafter referred to as “product position”) conveyed by the target equipment, and is set in advance. The set value is indicated by a one-dot chain line, and the actually measured value is indicated by a solid line. On the other hand, the time-series data in FIG. 2B is time-series data of the speed of the drive motor that conveys the steel product (hereinafter referred to as “motor speed”), and the set value set in advance is a one-dot chain line. The measured values are indicated by solid lines.

As described above, the target equipment repeatedly performs a fixed operation, that is, an operation of transporting the steel product to a predetermined transport position, and each time-series data of the product position and motor speed obtained from the target equipment. The change curve shows regularity (pattern) as shown in FIGS. 2A and 2B for each operation unit of the target equipment to be repeated. That is, the change curve of the product position shows the pattern of FIG. 2A for each operation unit of the target equipment, and the change curve of the time-series data of the motor speed shows the change curve of FIG. 2B for each operation unit of the target equipment. Indicates a pattern. In the present embodiment, the time series data of the product position of this operation unit is sampled at L sampling times (1, 2,..., L), and sampling data (sampled L time series data) D ₁₁ , D ₁₂ ,... D _1L are acquired as variables (variable A) representing the product position pattern. Similarly, the time series data of the motor speed of the operation unit is sampled at the same L sampling times (1, 2,..., L) as the variable A, and the sampling data (L time series data sampled) D ₂₁ , D ₂₂ ,... D _2L are acquired as variables (variable B) representing the motor speed pattern. Note that the sampling number L may be a fixed value set in advance or may be variably set in accordance with a user operation. The sampling number L may be set as appropriate. Can be obtained.

  If the variable A acquired by sampling L time-series data of product positions in this way is plotted in the L-dimensional space, the pattern of the product position represented by the variable A is represented by one point in the L-dimensional space. Can do. FIG. 3 is a diagram illustrating a distribution example of product position patterns in the L-dimensional space. If the operation of the target facility is normal, the obtained product position pattern follows a similar locus. Therefore, when a plurality of points P31 representing normal product position patterns are plotted in the L-dimensional space, each point P31 is distributed in a spherical shape in the L-dimensional space as shown by being surrounded by a one-dot chain line in FIG. On the other hand, when the operation of the target facility is abnormal, the point P33 representing the product position pattern deviates from the spherical distribution. Although not shown, the motor speed pattern similarly draws a similar locus if the operation of the target equipment is normal. When this motor speed pattern is expressed as a point in the L-dimensional space, what is obtained during normal operation is distributed spherically, and what is obtained during abnormal operation deviates from this spherical distribution.

  Here, a combination of the product position pattern and the motor speed pattern in the time axis direction is defined as one operation pattern of the target equipment. In this case, a variable A acquired by sampling L time-series data of product positions and a variable B acquired by sampling L time-series data of motor speeds as 2L variables in a 2L-dimensional space. If plotted, the motion pattern can be represented by one point in the 2L-dimensional space. FIG. 4 is a diagram illustrating an example of a motion pattern distribution in a 2L-dimensional space.

  As described above, since the operation pattern is a combination of the product position and motor speed patterns, similar to the product position and motor speed patterns, if the operation of the target equipment is normal, it is the same. Draw a trajectory. By the way, since the change in the product position is caused by the change in the motor speed, the variable A obtained by sampling the L time series data of the product position and the L time series data of the motor speed are sampled. It correlates with the obtained variable B. Therefore, when a plurality of points P41 representing a normal operation pattern (normal pattern) are plotted, the sampling data acquired as the variables A and B are elliptically shaped as indicated by the dotted lines in FIG. Distributed. On the other hand, the point P43 in the 2L-dimensional space that represents an abnormal operation pattern (abnormal pattern) when an abnormality such as hunting or response delay occurs in the target facility deviates from the elliptical distribution. In particular, when the correlation between a plurality of normal patterns is strong, the main component (ellipse major axis) appears remarkably. At this time, the abnormal pattern deviates from the elliptical distribution by increasing the out-of-phase component orthogonal to the main component (ellipse major axis) of the normal pattern.

  In this embodiment, the degree of dispersion is calculated in advance for a plurality of normal patterns in which the plot positions in the 2L-dimensional space are distributed elliptically, and the normal operating state of the target equipment is indexed. deep. Then, the state of the target equipment is monitored based on the degree of deviation from the distribution (degree of dispersion) of the normal pattern, and an abnormality of the target equipment is detected.

  As a method of indexing the degree of variance of normal patterns, principal component analysis is applied when the correlation is strong, and the average and variance are calculated when the correlation is weak. I decided to. For example, when the maximum value of the correlation coefficient between each variable of the normal pattern is equal to or greater than a predetermined threshold, principal component analysis is performed using this target facility as the second model, and the major principal component (the major axis of the ellipse). ), The normal operating state of the target equipment is indexed. Then, the abnormality of the target equipment is detected based on the deviation component orthogonal to the main main component as the deviation degree.

  When the maximum value of the correlation coefficient between each variable of the normal pattern is less than a predetermined threshold, the target equipment is used as the first model, and the average (center of gravity) and variance of each variable are calculated, and the target Index the normal operating state of the equipment. Then, based on the known Mahalanobis distance (the distance from the center of gravity divided by the width of the ellipse in that direction) as the degree of departure, an abnormality in the target facility is detected.

  In addition, here, a case where two types of variables A and B corresponding to two types of time series data of the product position and the motor speed of the target facility are acquired and the normal operation state of the target facility is indexed is illustrated. On the other hand, when acquiring M × L variables corresponding to the number of sampling data L from three or more types of M time-series data, the normal pattern is similarly expressed as M × L of each variable M × L. By expressing it as one point in the L-dimensional space, the normal operating state of the target facility is indexed. Similarly, the state of the target equipment is monitored based on the degree of deviation from the distribution of the normal pattern, and an abnormality of the target equipment is detected.

  Next, a specific processing procedure performed by the abnormality monitoring system 1 will be described. FIG. 5 is a flowchart showing a processing procedure of monitoring index creation processing performed by the offline index creation system 2. FIG. 6 is a flowchart showing a processing procedure of abnormality monitoring processing performed by the online diagnosis system 3. In the abnormality monitoring system 1, the offline index creation system 2 performs the monitoring index creation process according to the processing procedure of FIG. 5, and the online diagnostic system 3 performs the abnormality monitoring process according to the processing procedure of FIG. In the process described here, a program for realizing the monitoring index creation process is stored in the storage unit 23 of the offline index creation system 2, and the offline index creation system 2 reads and executes the program. A program for realizing the abnormality monitoring process is stored in the storage unit 33 of the online diagnostic system 3, and the online diagnostic system 3 reads out and executes this program.

  In the monitoring index creation process performed by the offline index creation system 2, as shown in FIG. 5, first, in the control unit 25, the partial time series cut-out processing unit 271 of the monitoring index creation processing unit 27 refers to the operation DB 4, For each of two or more types of M time series data (in this embodiment, two types of time series data of product position and motor speed) obtained from the target equipment at the time of past operation (at the target equipment) Time series data obtained during normal operation) is read (step S101). Then, the partial time-series cut-out processing unit 271 cuts out a plurality of each of the M types of time-series data at the normal operation read out in step S101 in units of operation of the target equipment (step S103). For example, the time length required for one operation of the target equipment to be repeated is set in advance. In step S103, the partial time-series extraction processing unit 271 extracts a plurality of time-series data having the above-described time length from each of the M types of time-series data with reference to the operation start position of the target facility.

  Subsequently, the normalization processing unit 273 performs normalization processing as preprocessing for the M types of time-series data cut out in step S103 (step S105). As described above, the target facility repeatedly performs the operation of sequentially transporting the steel product to be transported to the transport position corresponding to the type or the like. The normalization process in step S105 is for performing the subsequent processes after step S107 without being affected by the magnitude of the M types of time-series data resulting from the transport position.

  Subsequently, as a normal pattern extraction process, the statistical monitoring indexing processing unit 275 obtains M types of time series data (variable A and variable B in this embodiment) after the normalization processing in step S105. Each of the L samples is sampled, and the corresponding M × L types of variables are acquired (step S107). Then, the statistical monitoring indexing processing unit 275 extracts a plurality of normal patterns and calculates a correlation coefficient between each variable (step S108). When the maximum value of the correlation coefficient is equal to or greater than a predetermined threshold, the statistical monitoring indexing processing unit 275 determines that the target facility is the second model, and the monitoring index creation processing moves to the processing of step S109. On the other hand, if the maximum value of the correlation coefficient is less than the predetermined threshold, the statistical monitoring indexing processing unit 275 determines that the target facility is the first model, and the monitoring index creation processing moves to the processing of step S110. .

  In the process of step S109, the statistical monitoring indexing processing unit 275, as the monitoring index creation process, stores the M × L type variables acquired in step S107 in the M × L dimensional space (2L dimensional space in the present embodiment). A principal component analysis of a plurality of normal patterns expressed as points is performed (step S109).

  For example, in the present embodiment, a known principal component analysis is performed on a plurality of normal patterns represented by the points P41 that are distributed elliptically in the 2L-dimensional space shown in FIG. In this principal component analysis, using the sampling data as variables A and B of each normal pattern, eigenvectors for each principal component are obtained, conversion coefficients for each principal component are obtained, and principal component expressions are obtained. Next, the number k of components of the upper principal component, which is the main principal component, whose cumulative contribution rate is equal to or greater than a preset threshold value (for example, 0.8) is determined. The threshold for determining the number of components k may be a fixed value, may be set variably according to a user operation, and may be set as appropriate. The main characteristics that characterize the normal pattern are determined by the top k principal components (first to k-th principal components) determined here. On the other hand, an outlier component (residual), which is a principal component lower than the (k + 1) th principal component, has a low contribution when determining the characteristics of the normal pattern.

  When the principal component analysis is performed as described above and the principal component expression and the number k of the higher principal components expressed by the conversion coefficient of the principal component of the normal pattern are obtained, the statistical monitoring indexing processing unit 275 is obtained. Stores the obtained principal component formula and the number k of components in the second model monitoring index file 52 of the monitoring index DB 5 as a monitoring index obtained by indexing the normal operating state of the target equipment of the second model (step S111).

  On the other hand, in the process of step S110, the statistical monitoring indexing processing unit 275 uses the M × L type variable acquired in step S107 as the monitoring index creating step in the M × L dimensional space (in this embodiment, the 2L dimensional space). ) Calculate the average (center of gravity) and variance (average of deviation) of each variable of a plurality of normal patterns expressed as points in ().

  Then, the statistical monitoring indexing processing unit 275 uses the calculated average value and variance value as the monitoring index obtained by indexing the normal operating state of the target equipment of the first model, in the first model monitoring index file 51 of the monitoring index DB 5. (Step S112).

  In the abnormality monitoring process performed by the online diagnosis system 3, first, as shown in FIG. 6, in the abnormality monitoring processing unit 37 of the control unit 35, the partial time series cut-out processing unit 371 is obtained from the target equipment in operation. The received M types of time series data (in this embodiment, two types of time series data of product position and motor speed) are input (step S201), and each of the input M types of time series data is converted into the target equipment. The operation unit is cut out (step S203). In this process, as in step S101 of FIG. 5, the time length required for one operation of the target equipment to be repeated is set in advance, and the time length described above is obtained from each of the M types of time-series data. This is done by cutting out the series data. The extracted M types of time-series data are temporarily stored in the storage unit 33. After that, the normalization processing unit 373 performs normalization processing as preprocessing for the M types of time-series data temporarily stored (step S205).

  Subsequently, when the target facility is the second model (step S206, Yes), the abnormality monitoring process proceeds to the process of step S207. On the other hand, when the target facility is the first model (step S206, No), the abnormality monitoring process proceeds to the process of step S208.

  In the process of step S207, the statistical monitoring indexing processing unit 375 refers to the second model monitoring index file 52 of the monitoring index DB 5, and is represented by the monitoring index of the target facility, that is, the conversion factor of the main component of the normal pattern. The principal component equation and the number k of the higher principal components are read out (step S207). Subsequently, the statistical monitoring indexing processing unit 375 performs time series data of M types (two types of variable A and variable B in the present embodiment) after the normalization processing in step S205 as the operation time pattern extraction step. Are sampled L at a time, and the corresponding M × L variables are acquired (step S209). The statistical monitoring indexing processing unit 375 then calculates the M × L variable obtained in step S209 as a point in the M × L dimensional space (2L dimensional space in the present embodiment) as the operation deviation calculation step. By applying the principal component expression read out in step S207 to the operation pattern (operation pattern of the target equipment in operation) expressed as, the operation pattern, that is, the sampling acquired as M × L type variables in step S209 The k principal components of the data and the outlier component which is a principal component lower than the k principal components are calculated (step S211).

  Thereafter, as a determination step, the determination processing unit 377 calculates a Q statistic of the outlier component as a deviation from the main component, and determines an abnormality of the target facility based on the deviation as the degree of deviation (step S213). . For example, the determination processing unit 377 determines the magnitude of the departure degree by performing threshold processing on the departure degree using a preset threshold value. Then, the determination processing unit 377 determines that the operation state of the target facility is abnormal when the deviation degree is large. The determination processing unit 377 counts the number of times determined to be abnormal here, and thresholds the number of counts in a predetermined period with a predetermined threshold value, thereby determining whether or not the target facility needs to be repaired (step S215).

  On the other hand, in the process of step S208, the statistical monitoring indexing processing unit 375 refers to the first model monitoring index file 51 of the monitoring index DB 5, and obtains the monitoring index of the target facility, that is, the normal pattern average and variance values. read out. Subsequently, the statistical monitoring indexing processing unit 375 performs time series data of M types (two types of variable A and variable B in the present embodiment) after the normalization processing in step S205 as the operation time pattern extraction step. Are sampled L at a time, and the corresponding M × L variables are obtained (step S210). The statistical monitoring indexing processing unit 375 then calculates the M × L variable obtained in step S210 as a point in the M × L dimensional space (2L dimensional space in the present embodiment) as the operation deviation calculation step. The Mahalanobis distance is calculated based on the average (center of gravity) and variance values read out in step S208 for the operating pattern (operation pattern of the target facility in operation) expressed as (step S212).

  Thereafter, as a determination step, the determination processing unit 377 determines abnormality of the target facility based on the deviation degree, using the calculated Mahalanobis distance as a deviation degree (step S213). For example, as described above, the determination processing unit 377 determines the degree of deviation by threshold processing the deviation using a preset threshold. Then, the determination processing unit 377 determines that the operation state of the target facility is abnormal when the deviation degree is large. The determination processing unit 377 counts the number of times determined to be abnormal here, and thresholds the number of counts in a predetermined period with a predetermined threshold value, thereby determining whether or not the target facility needs to be repaired (step S215).

  As described above, in the present embodiment, M × L types of sampling data sampled from each of the time-series data in the section in which one or more types of M predetermined operations obtained from the target facility are performed. It was decided to get it as a variable. Then, a point in the M × L dimensional space determined by the acquired M × L type variable is used as a normal pattern of the target facility, and the degree of dispersion of a plurality of normal patterns is calculated and used as a monitoring index, and the target facility at the time of operation The degree of deviation from the normal pattern distribution of the operating pattern was calculated to determine the abnormality of the target equipment.

  For example, when the correlation between the normal patterns is strong, the principal component analysis of the normal patterns is performed to obtain the principal component expression and the number k of the higher principal components represented by the transform coefficients of the principal components of the normal pattern. It was decided. Also, using the acquired principal component formula and the number k of the upper principal components as monitoring indicators, the principal component and outlier component of the operation pattern of the target equipment during operation are calculated, and the Q statistic of the outlier component deviates It was decided to determine the abnormality of the target equipment using it as a degree. When the correlation between normal patterns is weak, the average (center of gravity) and variance of normal patterns are calculated. Also, calculate the Mahalanobis distance for each point of the operating pattern of the target equipment during operation using the calculated average and variance values as monitoring indicators, and determine the abnormality of the target equipment using the Mahalanobis distance as the deviation It was.

  According to the present embodiment, it is possible to monitor the state of a general-purpose target facility without being aware of, for example, the type and characteristics of the target facility. That is, the state monitoring according to the present embodiment is not limited to the exemplified equipment, but is equipment that repeats a predetermined operation, and the equipment in which one or more types of M time series data obtained exhibit regularity in the unit of operation. If so, it can be applied universally regardless of the type of equipment. Therefore, for example, there is no need to change the parameter according to the type or characteristic of the target facility, or to reset the parameter according to the operating condition. For this reason, for example, the state monitoring in a large-scale plant with many plant facilities such as a steel product manufacturing plant can be simplified, and manpower and cost required for the state monitoring can be reduced. In addition, the M × L variable obtained by sampling M types of time-series data L at a time is used as the operation pattern of the target equipment, and the degree of dispersion of the normal pattern is used as a monitoring index to deviate from the operation time pattern. Since the degree can be determined, it is possible to realize highly accurate abnormality detection in consideration of the characteristics of the operation pattern of the target equipment (for example, the product position and motor speed pattern).

  In addition, although this Embodiment demonstrated the case where the abnormality was detected for the plant equipment of the steel product manufacturing plant as a monitoring object, it is not limited to the steel product manufacturing plant, but desired monitoring of various product manufacturing plant equipment, etc. The present invention can be similarly applied when detecting an abnormality based on time-series data obtained from an object.

(Example)
According to the processing procedure shown in FIG. 5 and FIG. 6, the principal component analysis is performed on the normal pattern of the target equipment of the second model obtained in advance in the past operation, etc. The expression of the principal component expressed and the number k of components of the upper principal component were obtained. Then, the state of the target equipment of the second model can be monitored by calculating the main components of the operating pattern based on the time series data of the product position and motor speed for 12 times obtained from the target equipment at the time of operation. went. 7-12 is a figure explaining the example of application of above-described embodiment.

  Here, FIG. 7A is a diagram in which product position patterns (patterns of variable A) of 12 times of each operation are overlapped, and FIG. 7B is a diagram of motor speeds of 12 times of each operation. It is the figure which piled up and showed the pattern (pattern of the variable B). FIG. 8 shows only the product position pattern (a) and the motor speed pattern (b) taken out during the abnormal operation of the target equipment out of the 12 product position and motor speed patterns. It is a figure. 9 (a), 9 (b),... Are among the operation time patterns in which the product position and motor speed patterns of 12 operations are combined in the time axis direction during normal operation of the target equipment. It is a figure which shows the pattern at the time of operation (normal pattern (1), (2), ...). FIG. 10 shows an operation pattern (abnormal pattern) during an abnormal operation among the 12 operations, specifically, the product position and motor speed patterns shown in FIGS. 8 (a) and 8 (b). It is a figure which shows the pattern at the time of operation couple | bonded with the time-axis direction. In FIGS. 8 and 10, the portion surrounded by the alternate long and short dash line in FIGS. 8 and 10A of the pattern of the product position from which the variable A is acquired is wavy due to the occurrence of hunting in the target equipment. It is a trajectory.

  FIG. 11 is a diagram illustrating the calculation result of the principal component performed by applying the principal component expression represented by the conversion coefficient of the principal component of the normal pattern to the operation pattern of each of 12 operations. For example, the top five principal components of the 52 principal components calculated with the sampling data number L = 26 and the component number k = 5 are quantified and shown every 12 operations. On the other hand, FIG. 12 is a diagram showing Q statistics calculated as deviations of outlier components other than the top five principal components. Here, of the 12 operations, the 10th operation corresponds to the abnormal operation shown in FIG. 8 and FIG. 10, and the 10th Q statistic is shown by being surrounded by a one-dot chain line in FIG. The (deviance) is sufficiently large compared to other Q statistics. Therefore, the tenth operation can be determined as abnormal from the value of the Q statistic.

  As described above, the abnormality monitoring system and abnormality monitoring method of the present invention are suitable for general-purpose application without being aware of the type of monitoring target and the like and detecting the abnormality with high accuracy.

DESCRIPTION OF SYMBOLS 1 Abnormality monitoring system 2 Offline index production system 21 Input part 22 Display part 23 Storage part 25 Control part 27 Monitoring index creation process part 271 Partial time series cut-out process part 273 Normalization process part 275 Statistical monitoring indexation process part 3 Online Diagnosis system 31 Input unit 32 Display unit 33 Storage unit 35 Control unit 37 Abnormality monitoring processing unit 371 Partial time series extraction processing unit 373 Normalization processing unit 375 Statistical monitoring indexing processing unit 377 Judgment processing unit 4 Operation DB
5 Monitoring index DB
51 First model monitoring index file 52 Second model monitoring index file

Claims (4)

An abnormality monitoring system for detecting an abnormality in the constant operation facility based on time series data obtained from a constant operation facility that repeatedly performs a predetermined operation during normal operation in a manufacturing process,
L times are sampled from each of the time series data of the section where the predetermined operation of M types (M ≧ 1) obtained from the constant operation equipment during normal operation is performed in advance, and acquired as M × L type variables. Normal pattern extraction means for setting a point in the M × L dimensional space determined by the M × L types of variables as a normal pattern and calculating a correlation coefficient between each variable of the plurality of normal patterns in the space;
When the maximum value of the correlation coefficient is less than a predetermined threshold, the average and variance of each variable are calculated using the constant operation facility as the first model, and the maximum value of the correlation coefficient is equal to or greater than the predetermined threshold A monitoring index creating means for calculating principal component conversion coefficients by performing principal component analysis on each variable using the constant operation facility as a second model;
The M × L types of variables are obtained from each of the time-series data of the section in which the M types of predetermined operations are performed, which are obtained from the constant operation facility during operation, and M × L determined by the M × L types of variables An operation time pattern extraction means for making a point in the L-dimensional space an operation time pattern;
When the constant operation facility is the first model, the Mahalanobis distance of the operation pattern is calculated as a deviation based on the average and variance of the normal patterns acquired by the monitoring index creating means, and the constant operation facility When the second model is the second model, the deviation from the main component of the operating pattern is calculated as the deviation based on the conversion factor of the main component of the normal pattern acquired by the monitoring index creating means. Means,
A determination means for determining an abnormality of the constant operation facility based on the deviation calculated by the operating deviation calculation means;
An abnormality monitoring system comprising:   2. The abnormality monitoring system according to claim 1, wherein the determination unit determines whether or not the constant operation equipment needs to be repaired based on the number of times that the constant operation equipment is determined to be abnormal in a predetermined period. . The monitoring index creating means further determines a main principal component by calculating a contribution ratio of the main component of the normal pattern for the constant motion facility of the second model,
The determination means calculates the deviation based on a deviation component other than the main principal component among the principal components of the operating pattern, and determines that the constant operation facility is abnormal when the deviation is a predetermined value or more. The abnormality monitoring system according to claim 1, wherein the abnormality monitoring system is determined. An abnormality monitoring system constructed by a computer, wherein the computer detects an abnormality of the constant operation facility based on time series data obtained from a constant operation facility that repeatedly performs a predetermined operation during a normal operation in a manufacturing process. A method,
The computer samples in advance L pieces of time-series data of sections in which predetermined operations of M types (M ≧ 1) obtained from the constant operation equipment during normal operation are performed, and M × L types A normal pattern extraction step of obtaining a variable, calculating a correlation coefficient between each variable of a plurality of normal patterns in the space, using a point in the M × L dimensional space determined by the M × L types of variables as a normal pattern When,
When the maximum value of the correlation coefficient is less than a predetermined threshold , the computer calculates an average and a variance of each variable using the constant operation facility as a first model, and the maximum value of the correlation coefficient is a predetermined value In the case of the threshold value or more, a monitoring index creation step of calculating a principal component conversion coefficient by performing principal component analysis on each variable using the constant operation facility as a second model;
The computer acquires the M × L variables from each of the time-series data of the section in which the M predetermined operations obtained from the fixed operation facility are performed during operation, and the M × L variables An operation time pattern extraction step in which a point in the M × L dimensional space determined by
The computer calculates the Mahalanobis distance of the operating pattern as a deviation based on the average and variance of the normal pattern acquired in the monitoring index creation step when the constant operation facility is the first model, An operation for calculating a deviation from the main component of the operating pattern as a deviation based on the conversion coefficient of the main component of the normal pattern acquired in the monitoring index creation step when the constant operation facility is the second model Time deviation calculation process;
A determination step of determining abnormality of the constant operation equipment based on the deviation calculated by the computer- based deviation calculation step during operation;
An abnormality monitoring method comprising:

Images