JP2011247695A - Abnormality diagnosis method and abnormality diagnosis system using pattern library - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an abnormality diagnosis method and an abnormality diagnosis system using a pattern library, which can detect at high probability abnormality having no precedent in the past or abnormality having less precedent in the past.SOLUTION: An abnormality diagnosis system 1 using a pattern library, which detects abnormality in an abnormality diagnosis target based on time series data obtained from the abnormality diagnosis target includes: a normal pattern library 42 which stores normal patterns to be a time series patterns of a normal state previously acquired from the abnormality diagnosis target; and a normal pattern collation processing part 24 which collates the time series data with normal patterns, and when the time series data coincide with at least one normal pattern, determines that the abnormality diagnosis target is normal in a section from which the time series data are acquired.

Description

  The present invention relates to an abnormality diagnosis method and an abnormality diagnosis system using a pattern library, and in particular, monitors the state of mechanical equipment and processes based on time series data such as acoustic data, vibration data, and temperature data obtained from a steel plant. The present invention relates to an abnormality diagnosis method and an abnormality diagnosis system using a pattern library for performing the above.

  Conventionally, many techniques relating to abnormality diagnosis of mechanical equipment and process operation are known. As an example, there is a technique described in Patent Document 1. This is a technique for diagnosing an abnormality by registering in advance a typical pattern seen at the time of occurrence of an abnormality with respect to time-series data obtained from a process, and collating it with this. In other words, it is a technique based on the premise that a tendency at the time of occurrence of an abnormality is included as prior knowledge.

  Moreover, there are vibration diagnosis and acoustic diagnosis as practical methods for diagnosing abnormalities in mechanical equipment. For example, there is a technique described in Patent Document 2 for acoustic diagnosis. This is an abnormality diagnosis technique for the plunger pump, and since a change in sound at the time of an abnormality is known in advance, it is a technique for capturing the change. That is, the technique described in Patent Document 2 is also a technique based on the premise that the tendency at the time of occurrence of abnormality is included as prior knowledge.

JP-A-8-221113 Japanese Patent Laid-Open No. 2001-324381

  However, since the above-described conventional technologies are all premised on having a tendency at the time of occurrence of an abnormality as prior knowledge, there is a problem in that an unprecedented abnormality cannot be detected in the past. In addition, even if there are precedents of abnormalities in the past, if there are only a few cases, it is often difficult to have a tendency at the time of occurrence of abnormalities as prior knowledge, and in this case, the problem of overlooking detection of abnormalities There was a point.

  For example, in the steel process, there are many serious troubles that have to stop the factory for a long time once it occurs. However, it is not uncommon for such serious troubles to have no precedent in the past, or to have only a few. In such a process, there are no or very few cases and data at the time of abnormality, and the conventional approach based on past abnormality cases as described in Patent Document 1 misses the occurrence of abnormality. There is a limit.

  The present invention has been made in view of the above, and an abnormality using a pattern library that can detect this abnormality with a high probability even if there is an abnormality that has not been preceded in the past or an abnormality that has been unprecedented in the past. An object is to provide a diagnosis method and an abnormality diagnosis system.

  In order to solve the above-described problems and achieve the object, the abnormality diagnosis method using the pattern library according to the present invention detects an abnormality of the abnormality diagnosis target based on time series data obtained from the abnormality diagnosis target. An abnormality diagnosis method using a pattern library that stores a normal pattern that is a time series pattern in a normal state acquired in advance from the abnormality diagnosis target, and compares the time series data with the normal pattern. When the time series data matches at least one normal pattern, the abnormality diagnosis target in the section in which the time series data is acquired is diagnosed as normal.

  Further, the abnormality diagnosis method using the pattern library according to the present invention is the above invention, wherein the time-series data is compared with the normal pattern, and if the time-series data does not match all the normal patterns, The abnormality diagnosis target in the section in which the series data is acquired is diagnosed as having a possibility of abnormality.

  In the abnormality diagnosis method using the pattern library according to the present invention, in the above invention, the time series data diagnosed as having the possibility of abnormality is stored in a temporary storage library, and then the time series is stored. An abnormality diagnosis is performed by performing detailed collation processing on the data by an external device.

  The abnormality diagnosis method using the pattern library according to the present invention is characterized in that, in the above invention, when the signal amplitude level of the time series data is a predetermined value or less, the time series data is diagnosed as normal. And

  Further, the abnormality diagnosis method using the pattern library according to the present invention is characterized in that, in the above invention, the matching determination between the time series data and the normal pattern is performed based on a pattern matching degree.

  In the abnormality diagnosis method using the pattern library according to the present invention, in the above invention, an abnormality pattern that is a time series pattern of an abnormal state acquired in advance from the abnormality diagnosis target is stored, and the time series data is stored. Before matching with the normal pattern, the time series data and the abnormal pattern are matched, and when the time series data and the abnormal pattern do not match, the time series data and the normal pattern are matched. And performing an abnormality diagnosis.

  An abnormality diagnosis system using the pattern library according to the present invention is an abnormality diagnosis system using a pattern library for detecting an abnormality of the abnormality diagnosis target based on time series data obtained from the abnormality diagnosis target. A normal pattern library that stores a normal pattern that is a time-series pattern in a normal state acquired in advance from the abnormality diagnosis target, the time-series data and the normal pattern are collated, and the time-series is converted into at least one normal pattern. A normal pattern matching processing unit for diagnosing that the abnormality diagnosis target in the section in which the time series data is acquired is normal when the data matches.

  Further, in the abnormality diagnosis system using the pattern library according to the present invention, in the above invention, the normal pattern collation processing unit collates the time-series data with the normal pattern, and the time-series data for all normal patterns. If the data do not match, the abnormality diagnosis target in the section in which the time series data is acquired is diagnosed as having a possibility of abnormality.

  In the abnormality diagnosis system using the pattern library according to the present invention, in the above invention, when the signal amplitude level of the time series data is a predetermined value or less, the noise determination process for diagnosing that the time series data is normal It has the part.

  In the present invention, an abnormality diagnosis method using a pattern library for detecting an abnormality of the abnormality diagnosis target based on time series data obtained from the abnormality diagnosis object, the normal state acquired in advance from the abnormality diagnosis object A normal pattern that is a time series pattern of the time series data is stored, the time series data is compared with the normal pattern, and when the time series data matches at least one normal pattern, the time series data is acquired. Since the abnormality diagnosis target in the section is diagnosed as being normal, even if there is an unprecedented abnormality in the past or an abnormality with few precedents in the past, this abnormality can be detected with high probability.

FIG. 1 is a functional block diagram showing a configuration of an abnormality diagnosis system according to an embodiment of the present invention. FIG. 2 is a flowchart showing an abnormality diagnosis processing procedure by the control unit. FIG. 3 is a flowchart showing the temporary library determination processing procedure by the external device. FIG. 4 is a waveform diagram showing a normal pattern a. FIG. 5 is a waveform diagram showing a normal pattern b. FIG. 6 is a waveform diagram showing a normal pattern c. FIG. 7 is a waveform diagram showing an example of time-series data obtained from an abnormality diagnosis target and a noise level determination result. FIG. 8 is a waveform diagram showing the degree of pattern matching between the time series data shown in FIG. 7 and the normal patterns a, b, and c and the overall determination result. FIG. 9 is a waveform diagram showing the degree of pattern matching between the time series data shown in FIG. 7 and the normal patterns a and b and the overall determination result. FIG. 10 is a waveform diagram showing a normal pattern d. FIG. 11 is a diagram showing the pattern matching degree between the time series data and the normal pattern d and the temporal change of the time average. FIG. 12 is a diagram showing a relationship between a normal sound pattern and its generation timing.

  Embodiments of an abnormality diagnosis method and an abnormality diagnosis system using a pattern library according to the present invention will be described below with reference to the drawings. Note that the present invention is not limited to the embodiments.

  FIG. 1 is a functional block diagram showing a configuration of an abnormality diagnosis system according to an embodiment of the present invention. In FIG. 1, in this abnormality diagnosis system 1, an input unit 11, an output unit 12, and a storage unit 13 are connected to a control unit C including an abnormality diagnosis processing unit 20. The control unit C is realized by a CPU or the like, the input unit 11 is realized by a data collection device, a pointing device, or the like, the output unit 12 is realized by a liquid crystal display or the like, and the storage unit 13 is realized by a hard disk device or the like. The An external device 14 is also connected to the control unit C, and communication with the control unit C is possible via a communication interface.

  The storage unit 13 includes an abnormal pattern library 41 that stores abnormal patterns at the time of past abnormalities, a normal pattern library 42 that stores a large number of normal patterns at normal times, and a plurality of patterns that are unknown as normal patterns or abnormal patterns. And a temporary library 43 in which is stored. The abnormal pattern library 41 stores an abnormal pattern when an abnormal pattern is obtained, and does not store the abnormal pattern when there is no abnormality and the abnormal pattern cannot be obtained, and remains empty.

  The abnormality diagnosis processing unit 20 includes a cut-out processing unit 21, a noise determination processing unit 22, an abnormal pattern matching processing unit 23, a normal pattern matching processing unit 24, a registration processing unit 25, and an alarm output processing unit 26. I do.

  Here, the abnormality diagnosis processing procedure by the abnormality diagnosis processing unit 20 will be described with reference to the flowchart shown in FIG. In FIG. 2, for example, when time series data of sound pressure acquired by a sound collector installed near a mechanical facility such as a rolling mill in a steel process is sequentially input from the input unit 11, the cutting processing unit 21 A process of cutting out the data as a partial sequence that is time-series data having a preset time length is performed (step S101). This time length is set to the longest time length among the patterns stored in the abnormal pattern library 41 or the normal pattern library 42.

  Thereafter, the noise determination processing unit 22 determines whether or not the signal amplitude level of the partial sequence is equal to or lower than a predetermined value, that is, whether or not it is a noise level (step S102). When the signal amplitude level of the partial sequence is the noise level (Yes in step S102), the process proceeds to step S107 and is determined as “normal”.

  On the other hand, when the signal amplitude level of the partial sequence is not the noise level (No in step S102), the process proceeds to step S103, assuming that the pattern is a significant pattern including an abnormal pattern or a normal pattern.

  Thereafter, the abnormal pattern matching processing unit 23 performs an abnormal pattern matching process for matching the partial sequence cut out by the cutting processing unit 21 with the abnormal pattern in the abnormal pattern library 41 (step S103). This abnormal pattern matching process performs a matching process on all of the abnormal patterns stored in the abnormal pattern library 41 for the partial sequence. If the partial sequence matches at least one abnormal pattern, a step is performed. The process proceeds to S104. If no abnormal pattern is stored in the abnormal pattern library 41, the abnormal pattern matching process is skipped.

  As a result of the abnormal pattern matching process, it is determined whether or not there is an abnormal pattern that matches the extracted partial sequence (step S104). If they match (Yes in step S104), the process proceeds to step S108. It is determined as “abnormal”. When it is determined as “abnormal”, the warning output processing unit 26 outputs a warning to that effect from the output unit 12.

  On the other hand, when there is no abnormal pattern that matches the extracted partial sequence (No in step S104), the normal pattern matching processing unit 24 further converts the extracted partial sequence as a normal pattern in the normal pattern library 42. A normal pattern matching process for matching is performed (step S105). In the normal pattern matching process, as in the case of the abnormal pattern matching process, the matching process is performed on all the normal patterns stored in the normal pattern library 42, but the partial sequence matches at least one normal pattern. The process proceeds to step S106.

  As a result of the normal pattern matching process, it is determined whether or not there is a normal pattern that matches the extracted partial sequence (step S106). If they match (step S106, Yes), the process proceeds to step S107. It is determined as “normal”. If it is determined as “normal”, the alarm output processing unit 26 may output the fact from the output unit 12.

  On the other hand, if there is no normal pattern that matches the extracted partial sequence (No in step S106), the normal pattern matching processing unit 24 may have an error in the partial sequence pattern. It is determined as “abnormal”. Since the pattern of the subsequence is neither normal nor abnormal at this time, the alarm output processing unit 26 outputs an alarm from the output unit 12 to indicate that “the state is different from usual”. Further, the registration processing unit 25 registers the pattern of the partial sequence in the temporary placement library 43 (step S109). When registering in the temporary library 43, the registration processing unit 25 preferably registers the occurrence time data of this partial sequence in association with this partial sequence.

  Next, after the abnormality determination in step S108, the normality determination in step S107, or the temporary abnormality determination in step S109, it is determined whether or not there is an instruction to end this processing from the input unit 11 (step S110). When there is an end instruction (step S110, Yes), this process is ended. When there is no end instruction (step S110, No), the process proceeds to step S101, and the next partial sequence is cut out. Repeat the process.

  The external device 14 functions as a temporary placement library determination processing system, and performs detailed determination of whether the pattern of the partial sequence registered in the temporary storage library 43 is an abnormal pattern or a normal pattern. Temporary library determination processing registered in the pattern library 41 or the normal pattern library 42 is performed. The external device 14 has, for example, a database in which past operation data, unit test data of equipment, and the like are stored. Based on this database, a detailed verification process (detailed verification process) is performed on whether a partial sequence is an abnormal pattern or a normal pattern. ) Is realized by a large computer. Of course, determination processing by a person such as an expert may be added to the detailed collation processing by the external device 14.

  Here, with reference to the flowchart shown in FIG. 3, the temporary library determination processing procedure by the external apparatus 14 will be described. When there is an alarm notification from the output unit 12 of the abnormality diagnosis system 1 (Yes in step S201), the external device 14 accesses the temporary library 43 and extracts the registered partial sequence pattern (step S202). .

  Thereafter, the external device 14 performs detailed collation processing to determine whether the extracted partial sequence pattern is an abnormal pattern or a normal pattern (step S203). Thereafter, the external device 14 determines whether or not the pattern of the partial sequence on which the detailed matching process has been performed is abnormal (step S204). If it is determined that there is an abnormality (step S204, Yes), the registration processing unit 25 performs a process of registering the pattern of this partial sequence in the abnormality pattern library 41 (step S205). On the other hand, if it is not determined to be abnormal (No in step S204), the registration processing unit 25 performs a process of registering the pattern of this partial sequence in the normal pattern library 42 (step S206). In this way, the pattern of the “temporary abnormality” subsequence that is in an unusual state is determined to be abnormal or normal, and is stored by being registered in the abnormal pattern library 41 or the normal pattern library 42, respectively. By being used for the diagnosis process, a more accurate abnormality diagnosis process can be performed.

  Note that the temporary library determination process by the external device 14 may be processed online or may be processed offline with respect to the abnormality diagnosis system 1 as described above. For example, a “temporary abnormality” partial sequence pattern is periodically acquired from the temporary placement library 43 of the abnormality diagnosis system 1 and stored in a predetermined storage medium, and the stored partial sequence pattern is stored in the external device 14. By inputting, the external device 14 may perform the detailed collation process described above, and may be registered in the corresponding abnormal pattern library 41 or the normal pattern library 42 based on the collation process result.

(Verification processing example 1)
Here, an example of the abnormal pattern matching process in step S103 or the normal pattern matching process in step S105 will be described. In this collation processing example 1, the pattern matching degree R1 is calculated by correlating with the partial sequence of the time series data while sliding the abnormal pattern or the normal pattern by the following equation (1).
R1 = <S · P> / (<S · S> · <P · P>) (1)
Here, S is a cut out partial sequence, and P is a pattern registered in the abnormal pattern library 41 or the normal pattern library 42. <X · Y> indicates an inner product of X and Y. The pattern matching degree R1 becomes “1” when a partial sequence having the same pattern as the pattern registered in the abnormal pattern library 41 or the normal pattern library 42 appears. If they do not match, the value approaches “0”.

  An example of continuous abnormality diagnosis processing by normal pattern matching processing using the pattern matching degree R1 will be further described. In this abnormality diagnosis processing example, three normal patterns a, b, and c as shown in FIGS. 4 to 6 are stored in the normal pattern library 42, and the time series data as shown in FIG. Abnormality diagnosis processing is performed. The time series data shown in FIG. 7A is data composed of a plurality of partial columns. In the time-series data shown in FIG. 7A, one or more combinations of normal patterns a, b, and c are inherent at various locations on the time axis of 1000 seconds.

  8 (a) to 8 (c) show the correlation by sliding the normal patterns a, b, and c on the time axis with respect to the time series data shown in FIG. 7 (a). This is a result of calculating the pattern matching degree R1. Each of FIGS. 8A to 8C shows a temporal change in the pattern matching degree R1. For example, in the time-series data shown in FIG. 7A, the normal pattern a is included in the vicinity of 100 seconds, so the pattern matching degree R1 shown in FIG. 8A is close to “1” in the vicinity of 100 seconds. The peak value is shown.

  In FIG. 8A to FIG. 8C, in order to determine the match of the time series data pattern shown in FIG. 7A based on the pattern matching degree R1, for example, the pattern matching degree R1 is 0. In the case of .5 or more, “1” is output as matching, and “0” is output as not matching if less than 0.5. In this case, when at least one of the determination results in FIGS. 8A to 8C is “1” at a certain point in time, the time-series data is determined to match the normal pattern and determined to be normal. Is done. Further, even when all of the determination results in FIGS. 8A to 8C are “0” at a certain point in time, the noise level is not more than a predetermined value as shown in FIG. 7B. For example, when the signal amplitude level is 0.6 or less, “1” is output as a section where there is no effective pattern. As a result, when at least one of the determination results in FIGS. 8A to 8C and FIG. 7B is “1” as in the comprehensive determination result illustrated in FIG. When the logical sum operation is “1”, the time-series data at each time point is determined to be “normal”, and when the logical sum operation is “0”, the time-series data of each dictionary is the subsequent details. It is a target of the collation process, and is determined as “normal” or “abnormal”. In the comprehensive determination result of FIG. 8D, it is determined that “normal” in all the sections. Since the pattern matching degree R1 has a predetermined time length, the determination results in FIGS. 8A to 8C are sections in which the pattern length is increased.

  More specifically, when abnormality diagnosis processing is performed only on the normal patterns a and b for the time series data including the normal patterns a, b, and c in FIG. 7A, FIG. 9A and FIG. As shown in FIG. 9B, the same pattern matching degree R1 as in FIGS. 8A and 8B is obtained, but the pattern matching degree R1 corresponding to the pattern c corresponding to FIG. Therefore, even if the determination result of the noise level shown in FIG. 7B is obtained, the overall determination result is around 600 seconds (section E1) as shown in FIG. 9C. In the vicinity of 930 seconds (section E2), the logical sum is “0”. This is because the unexpected pattern appears in the sections E1 and E2 corresponding to the time point when the normal pattern c is included in the time series data shown in FIG. It will be done. Of course, as shown in FIG. 8, by performing the matching process on the normal pattern c, the overall determination result is determined to be “normal” in the sections E1 and E2, with “1”.

(Verification processing example 2)
Next, as another example of the normal pattern matching process, a pattern matching degree R2 shown in the following equation (2) is used, and a minute change abnormality diagnosis process is performed based on the rate of change of the pattern matching degree R2. Also good.
R2 = <S · P> / (<P · P> · <P · P>) (2)

  In this collation processing example 2, when only one normal pattern d shown in FIG. 10 is used and the pattern matching degree R2 shown in the equation (2) is obtained for the time series data shown in FIG. A time change of the pattern matching degree R2 as shown in FIG. If the time average of the predetermined time width of the pattern matching degree R2 is taken, it is as shown in FIG.

  Here, in the time-series data shown in FIG. 11A, the normal pattern d appears periodically until around 650 seconds, but after 650 seconds, a pattern in which the normal pattern d has changed appears. For this reason, the time average of the pattern coincidence degree R2 with respect to the pattern d shown in FIG. 11C is almost “1” until around 650 seconds, but after 650 seconds, particularly after 700 seconds, suddenly “1”. .2 ”. Therefore, the time average threshold value of the pattern matching degree R2 is set to “1.1”, and when this threshold value is exceeded, it can be determined that the time series data has changed from “normal” to “abnormal”. In the abnormality diagnosis process shown in FIG. 2, when the time average value of the pattern matching degree R2 at each time exceeds the threshold, it is determined that the partial sequence does not match the normal pattern, and the threshold is not exceeded. , It is determined that the partial sequence matches the normal pattern.

  In this comparative collation example 2, only the normal pattern is used, and whether the time series data is normal or abnormal is diagnosed based on the change in the pattern matching degree. In the comparative collation example 2, the pattern matching degree R1 shown in the comparative collating example 1 may be used, or in the comparative collating example 1, the pattern matching degree R2 may be used. Furthermore, the pattern matching degrees R1 and R2 are an example of the matching degree between the pattern and the partial sequence, and other pattern matching degree expressions may be used.

  In this embodiment, at least only the normal pattern is used to detect whether the time-series data is normal or abnormal. Therefore, even if there is an unprecedented abnormality or an abnormality with few precedents in the past, it is usually obtained. This abnormality can be detected with high probability only by the normal pattern. When an abnormal pattern is obtained, the abnormal pattern matching process is performed prior to the normal pattern matching process, whereby a quick and efficient abnormality diagnosis process can be performed.

  In addition, when performing the abnormality diagnosis by the sound for the rolling mill in the steel process, the normal pattern (sound pattern at normal time) shown in FIG. 12 can be used. These normal sound patterns can be obtained at the corresponding occurrence timings. For example, they can be collected individually during equipment unit tests conducted during regular repairs at the factory, or during operation. it can. By making this collected pattern into a library, it is possible to apply the present invention to perform abnormality diagnosis. For example, “rolling roll driving sound” as a normal sound pattern can be collected at the time of “rolling roll rotation” as the generation timing.

DESCRIPTION OF SYMBOLS 1 Abnormality diagnosis system 11 Input part 12 Output part 13 Memory | storage part 14 External apparatus 20 Abnormality diagnosis process part 21 Cutout process part 22 Noise determination process part 23 Abnormal pattern collation process part 24 Normal pattern collation process part 25 Registration process part 26 Alarm output process Unit 41 Abnormal pattern library 42 Normal pattern library 43 Temporary placement library C Control unit

Claims (9)

Based on the time series data obtained from the abnormality diagnosis target, an abnormality diagnosis method using a pattern library for detecting the abnormality of the abnormality diagnosis target,
A normal pattern that is a time series pattern in a normal state acquired in advance from the abnormality diagnosis target is stored, the time series data and the normal pattern are collated, and the time series data matches at least one normal pattern. In this case, the abnormality diagnosis method using the pattern library is characterized in that the abnormality diagnosis target in the section in which the time series data is acquired is diagnosed as normal.   The time-series data is compared with the normal pattern, and if the time-series data does not match all the normal patterns, the abnormality diagnosis target in the section in which the time-series data is acquired is diagnosed as possibly having an abnormality An abnormality diagnosis method using the pattern library according to claim 1. The time series data diagnosed as having the possibility of abnormality is stored in a temporary library, and then an abnormality diagnosis is performed by an external device performing detailed matching processing on the time series data. An abnormality diagnosis method using the pattern library according to claim 2.   The abnormality using the pattern library according to any one of claims 1 to 3, wherein when the signal amplitude level of the time series data is a predetermined value or less, the time series data is diagnosed as normal. Diagnosis method.   5. The abnormality diagnosis method using a pattern library according to claim 1, wherein the match determination between the time series data and the normal pattern is performed based on a pattern matching degree.   An abnormal pattern that is a time-series pattern of an abnormal state acquired in advance from the abnormality diagnosis target is stored, and the time-series data and the abnormal pattern are collated before the time-series data and the normal pattern are collated. 6. When the time-series data and the abnormal pattern do not match, the time-series data and the normal pattern are collated to perform an abnormality diagnosis. An abnormality diagnosis method using the pattern library according to one. Based on the time-series data obtained from the abnormality diagnosis target, an abnormality diagnosis system using a pattern library that detects the abnormality of the abnormality diagnosis target,
A normal pattern library storing normal patterns that are time series patterns in a normal state acquired in advance from the abnormality diagnosis target,
Normal that compares the time-series data with the normal pattern and, when the time-series data matches at least one normal pattern, diagnoses that the abnormality diagnosis target in the section in which the time-series data is acquired is normal A pattern matching processing unit;
An abnormality diagnosis system using a pattern library characterized by comprising:   The normal pattern collation processing unit collates the time series data with the normal pattern, and when the time series data does not match all the normal patterns, the abnormality diagnosis target in the section in which the time series data is acquired is 8. The abnormality diagnosis system using the pattern library according to claim 7, wherein diagnosis is made that there is a possibility of abnormality.   9. The pattern library according to claim 7, further comprising a noise determination processing unit that diagnoses that the time series data is normal when the signal amplitude level of the time series data is a predetermined value or less. Had an abnormal diagnosis system.

Images