JP2019185422A - Failure prediction method, failure prediction device, and failure prediction program - Google Patents

Abstract

A failure prediction system for improving failure prediction accuracy of mechanical equipment is provided. An extraction unit extracts a normal state, an abnormal state, and an evaluation at an arbitrary evaluation time from among sensor data of a plurality of sensors of a machine. The first generation unit 11c generates a normal model in which the correlation between sensors is modeled by machine learning using a normal state component. The second generation unit 11d generates, based on the normal model, an abnormal classification model in which an abnormal pattern is modeled by transfer learning using an abnormal state component. The evaluation unit 11e evaluates the degree of deviation from the normal state of the mechanical equipment by inputting the evaluation into the normal model. The determination unit 11f determines an abnormal pattern based on an output value obtained by inputting the above-described evaluation into the abnormality classification model when it is determined that the mechanical equipment has a failure sign based on the degree of deviation. The notification unit 11g notifies the determination result of the determination process including the abnormal pattern. [Selection] Figure 2

Description

  Embodiments disclosed herein relate to a failure prediction method, a failure prediction device, and a failure prediction program.

  Conventionally, a technique for predicting the occurrence of a failure by detecting a failure sign by monitoring the sensor value of a sensor provided in the machine facility is known (for example, see Patent Document 1).

  Conventionally, rule information including conditions for performing an intervention operation is generated by executing machine learning based on past intervention operation results for the machine equipment, and is performed based on the rule information and the current state of the machine equipment. A technique for determining and instructing an intervention operation to be performed is known (for example, see Patent Document 2). The aforementioned rule information corresponds to a so-called prediction model.

  Here, by applying the technique disclosed in Patent Document 2 to the technique disclosed in Patent Document 1, for example, by performing machine learning on the sensor value indicated by the sensor when a failure occurs, it is considered that failure prediction using a prediction model becomes possible. .

JP 2011-230634 A JP 2017-049801 A

  However, the above-described prior art has room for further improvement in improving the accuracy of failure prediction of mechanical equipment.

  Specifically, it is practically difficult to collect and cover machine learning data indicating the state of all failures at the initial stage of generating a prediction model. In particular, when the machine facility is a large mechatronics machine such as a large refrigerator or a plant, it can be said that the number of sensors provided is enormous, which makes it more difficult.

  Even if data can be collected, machine learning is performed, and a prediction model can be generated, if an unknown pattern still exists in the actual operation data, a false detection occurs due to the unknown pattern, and a failure occurs. There is a risk that it will not be possible to take measures against the true cause of

  One aspect of the embodiments has been made in view of the above, and an object thereof is to provide a failure prediction method, a failure prediction device, and a failure prediction program capable of improving the failure prediction accuracy of mechanical equipment.

  The failure prediction method according to an aspect of the embodiment includes an extraction process, a first generation process, a second generation process, an evaluation process, a determination process, and a notification process. The extraction step includes, among sensor data of a plurality of sensors provided in the mechanical equipment, a normal state portion during normal operation of the mechanical equipment, an abnormal state portion when an abnormality occurs, and an evaluation portion at an arbitrary evaluation time. Extract. The first generation step generates a normal model that models the correlation between the sensors during the normal operation by executing machine learning using the normal state. The second generation step generates an abnormality classification model that models an abnormal pattern when the abnormality occurs by executing transfer learning using the abnormal state based on the normal model. The evaluation step evaluates the degree of deviation from the normal state of the mechanical equipment based on the output value of the normal model obtained by inputting the evaluation portion to the normal model. The determination step is based on an output value of the abnormality classification model obtained by inputting the evaluation portion to the abnormality classification model when it is determined that there is a failure sign in the mechanical equipment based on the degree of deviation. The abnormal pattern is determined. The notification step notifies a determination result of the determination step including the abnormal pattern.

  According to one aspect of the embodiment, it is possible to improve the failure prediction accuracy of mechanical equipment.

FIG. 1A is a schematic explanatory diagram (part 1) of the failure prediction method according to the embodiment. FIG. 1B is a schematic explanatory diagram (part 2) of the failure prediction method according to the embodiment. FIG. 1C is a schematic explanatory diagram (part 3) of the failure prediction method according to the embodiment. FIG. 1D is a schematic explanatory diagram (part 4) of the failure prediction method according to the embodiment. FIG. 1E is a schematic explanatory diagram (part 5) of the failure prediction method according to the embodiment. FIG. 2 is a block diagram of the failure prediction system according to the embodiment. FIG. 3A is an explanatory diagram (part 1) of a correlation score. FIG. 3B is an explanatory diagram (part 2) of the correlation score. FIG. 4A is a diagram (part 1) illustrating a specific example of an abnormality determination result by the failure prediction apparatus according to the embodiment. FIG. 4B is a second diagram illustrating a specific example of the abnormality determination result by the failure prediction apparatus according to the embodiment. FIG. 4C is a diagram (No. 3) illustrating a specific example of the abnormality determination result by the failure prediction apparatus according to the embodiment. FIG. 4D is a diagram (part 4) illustrating a specific example of the abnormality determination result by the failure prediction apparatus according to the embodiment. FIG. 4E is a diagram (No. 5) illustrating a specific example of the abnormality determination result by the failure prediction apparatus according to the embodiment. FIG. 4F is a diagram (No. 6) illustrating a specific example of the abnormality determination result by the failure prediction apparatus according to the embodiment. FIG. 5A is a flowchart (part 1) illustrating a processing procedure executed by the failure prediction apparatus. FIG. 5B is a flowchart (part 2) illustrating a processing procedure executed by the failure prediction apparatus. FIG. 6 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the failure prediction apparatus.

  Hereinafter, embodiments of a failure prediction method, a failure prediction device, and a failure prediction program disclosed in the present application will be described in detail with reference to the accompanying drawings. In addition, this invention is not limited by embodiment shown below.

  In the following, the machine equipment that is the target of failure sign determination is referred to as “target machine 100”. The target machine 100 is assumed to be a large mechatronics machine such as a power plant.

  First, an outline of the failure prediction method according to the present embodiment will be described with reference to FIGS. 1A to 1E. 1A to 1E are schematic explanatory views (No. 1) to (No. 5) of a failure prediction method according to an embodiment.

  As illustrated in FIG. 1A, the target machine 100 includes a sensor group of sensors S-1 to Sn. The failure prediction method according to the present embodiment grasps the correlation between the sensors S-1 to Sn based on the sensor data from the sensor group, and the entire target machine 100 based on the change in the correlation. It is to grasp the change of the behavior.

  Specifically, as shown in FIG. 1B, in the failure prediction method according to the present embodiment, machine learning is performed using the sensor data of the sensors S-1 to Sn indicating the correlation of “normal state”. This is executed (step S1), and a normal model 12d is generated. In the present embodiment, deep learning is used as an algorithm for machine learning, as illustrated in FIG. 1B with the normal model 12d as an auto encoder. Since deep learning is publicly known, detailed description is omitted.

  Here, the “normal state portion” refers to a portion based on each sensor data of a predetermined period (hereinafter sometimes referred to as “normal period”) in which the target machine 100 is in a normal state in the initial stage of operation or the like. The normal state of the target machine 100 can be modeled by the normal model 12d generated based on the correlation between the sensor data for the normal state.

  In the failure prediction method according to the present embodiment, as shown in FIG. 1B, sensor data of the sensors S-1 to Sn indicating the correlation of “evaluation” is input to the normal model 12d. Then, the amount of deviation of the correlation is calculated from the output value (regression value) of the normal model 12d obtained as a result. Then, the degree of deviation from the normal state is evaluated based on the amount of deviation of the correlation (step S2).

  If the degree of deviation from the normal state is large, a failure of the target machine 100 can be predicted by indicating a failure sign. Here, “evaluation portion” refers to, for example, a portion based on real-time data output from the target machine 100.

  As described above, in the failure prediction method according to the present embodiment, the behavior of the entire target machine 100 is grasped by the correlation between the sensors S-1 to Sn, and the change in the behavior models the correlation of the normal state. It can be obtained from the output value of the normalized normal model 12d. A failure of the target machine 100 is predicted based on the degree of deviation from the normal state based on the output value. For example, if the degree of deviation from the normal state is equal to or greater than a predetermined determination threshold indicating that there is a sign of failure, the administrator is notified of the abnormality.

  Therefore, according to the failure prediction method according to the present embodiment, for example, all failure phenomena that are necessary if the failure prediction method uses a prediction model generated by machine learning based on sensor data at the time of failure occurrence. A complicated process such as implementation of modeling is not required. That is, according to the failure prediction method according to the present embodiment, a failure sign of the target machine 100 can be easily grasped.

  In the failure prediction method according to the present embodiment, as shown in FIG. 1A, machine learning including time variation of each sensor data, that is, time-series correlation, is included in a feature vector (hereinafter simply referred to as “vector”). This makes it possible to grasp a sign of a failure that appears in the time fluctuation of sensor data. Thereby, the failure sign of the target machine 100 can be accurately captured.

  Further, in the failure prediction method according to the present embodiment, not only the normal model 12d that models the correlation of the normal state, but also the abnormal pattern at the time of occurrence of the abnormality is modeled based on the sensor data at the time of the past occurrence of the abnormality. By using the abnormality classification model 12e that has been used together, it is possible to determine even the types of abnormality.

  Specifically, as shown in FIG. 1C, in the failure prediction method according to the present embodiment, the abnormality classification is based on the sensor data of the sensors S-1 to Sn that indicate the correlation of “abnormal state”. A model 12e is generated.

  Here, “abnormal condition” refers to the amount based on each sensor data when various abnormalities occurred in the past, and as shown in the figure, for example, “XX abnormal” or “XX abnormal” Each past abnormal record pattern is labeled and a classification class is set (see “Category ID” in the figure). With the abnormality classification model 12e generated based on the sensor data for the abnormal state, it is possible to model an abnormality pattern when the abnormality occurs in the target machine 100.

  Further, in the failure prediction method according to the present embodiment, the abnormality classification model 12e is generated by performing transfer learning using the learned normal model 12d (step S3). By executing transfer learning, that is, machine learning using a hidden layer in the normal model 12d, the abnormal classification model 12e can be generated at high speed and with high accuracy.

  For the abnormality classification model 12e generated in this way, in the failure prediction method according to the present embodiment, as shown in FIG. 1D, for example, a high divergence degree determined as indicating a failure sign based on the normal model 12d. Input the sensor data of “Evaluation”.

  Then, based on the output value of the abnormality classification model 12e obtained as a result, for example, “category ID” and “correlation score” shown in the figure, the type of abnormality is determined (step S4), and a new event is additionally learned. (Step S5).

  A new event corresponds to, for example, a low correlation score, in other words, a low degree of similarity to a learned abnormality pattern, and by performing additional learning in which a new abnormality classification class is set for the new event, The accuracy of the abnormality classification model 12e can be further improved.

  Whether or not additional learning is necessary may be automatically determined by the system according to, for example, a correlation score, or may be determined based on knowledge of a person such as an administrator or an operator of the target machine 100. Specific contents regarding steps S4 and S5 will be described later with reference to FIGS. 3A to 4F.

  Thereby, according to the failure prediction method according to the present embodiment, it is estimated not only whether it is normal or abnormal, but also what is the cause when it is determined that there is an abnormality, that is, the target machine 100 has a failure sign. be able to.

  In the failure prediction method according to the present embodiment, the contents including the estimation result are notified, for example, by a mail service to the administrator of the target machine 100 as shown in FIG. 1E. Such notification can include a “past case” similar to the current abnormality and a “correlation score”.

  The administrator of the target machine 100 is not only able to grasp the failure signs of the target machine 100 in advance by checking the contents of the notification, but also the contents of necessary maintenance, arrangement, repair and replacement of personnel and parts. Thus, it is possible to prevent the target machine 100 from being suddenly broken or abnormally stopped. That is, according to the failure prediction method according to the present embodiment, it is possible to improve the maintainability of the target machine 100 and thus the reliability.

  Hereinafter, the configuration of the failure prediction system 1 to which the above-described failure prediction method is applied will be described more specifically.

  FIG. 2 is a block diagram of the failure prediction system 1 according to the present embodiment. In FIG. 2, constituent elements necessary for describing the features of the present embodiment are represented by functional blocks, and descriptions of general constituent elements are omitted.

  In other words, each component illustrated in FIG. 2 is functionally conceptual and does not necessarily need to be physically configured as illustrated. For example, the specific form of distribution / integration of each functional block is not limited to the one shown in the figure, and all or a part thereof is functionally or physically distributed in arbitrary units according to various loads or usage conditions・ It can be integrated and configured.

  In the description with reference to FIG. 2, the description of the components already described so far may be simplified or omitted.

  As shown in FIG. 2, the failure prediction system 1 includes a failure prediction device 10 and a target machine 100. The failure prediction apparatus 10 and the target machine 100 are provided so as to be communicable via a network connection, and the failure prediction apparatus 10 is provided so that sensor data from the target machine 100 can be collected as appropriate.

  The failure prediction apparatus 10 includes a control unit 11 and a storage unit 12. The control unit 11 includes a collection unit 11a, an extraction unit 11b, a first generation unit 11c, a second generation unit 11d, an evaluation unit 11e, a determination unit 11f, a notification unit 11g, and an update unit 11h. .

  The storage unit 12 is a storage device such as a hard disk drive, data flash, non-volatile memory, or register, and includes collected data 12a, a normal learning data set 12b, an abnormal learning data set 12c, a normal model 12d, The classification model 12e, the evaluation data set 12f, the evaluation information 12g, the past history data 12h, and the additional learning data set 12i are stored.

  The control unit 11 performs overall control of the failure prediction apparatus 10. The collection unit 11a collects sensor data from the sensor group of the target machine 100 at a predetermined period, and stores it in the collection data 12a. The predetermined period to be collected may be about 15 minutes to 1 hour in order to detect a gradual change in behavior indicating a failure sign due to secular change or the like.

  The extraction unit 11b extracts a data set for the normal state from the collected data 12a based on the normal period set at the first operation, and stores the data set in the normal learning data set 12b. Further, the extraction unit 11b extracts a data set for the past abnormal state from the collected data 12a and stores the data set in the abnormality learning data set 12c.

  Further, the extraction unit 11b extracts a data set for evaluation from the collected data 12a and stores it in the evaluation data set 12f.

  The first generation unit 11c executes machine learning using the normal learning data set 12b to generate a normal model 12d. The second generation unit 11d performs transfer learning using the normal model 12d and the abnormality learning data set 12c to generate the abnormality classification model 12e.

  The evaluation unit 11e inputs the evaluation data set 12f extracted by the extraction unit 11b to the normal model 12d and receives an output result from the normal model 12d. And the evaluation part 11e calculates the various evaluation values for determining a failure sign based on the received output result, and stores it in the evaluation information 12g. The evaluation value includes the degree of deviation from the normal state and the contribution rate.

  Specifically, the evaluation unit 11e calculates the degree of deviation from the normal state based on the input and output error when the evaluation data set 12f is input to the normal model 12d, that is, the amount of correlation shift, and the evaluation information Store to 12g.

  In addition, the evaluation unit 11e acquires the contributions of the sensors S-1 to Sn from the normal model 12d. For example, each sensor S-1 is expressed by an expression “contribution rate i = contribution i / Σcontribution i”. ~ Contribution rate for each Sn is calculated. The evaluation unit 11e stores the calculated contribution rate in the evaluation information 12g.

  The determination unit 11f refers to the evaluation information 12g, determines that there is a failure sign when the deviation from the normal state is equal to or greater than a predetermined determination threshold, and notifies the evaluation unit 11e of the determination result. When the evaluation unit 11e receives the determination result that there is a failure sign from the determination unit 11f, the evaluation unit 11e inputs the corresponding data with the failure sign to the abnormality classification model 12e, and outputs the result of the abnormality classification model 12e, that is, the category described above. The ID and the correlation score are received and stored in the evaluation information 12g.

  Then, the determination unit 11f refers to the evaluation information 12g, notifies the notification unit 11g of the divergence degree, the contribution rate, the category ID, and the correlation score of the corresponding data, and notifies the notification unit 11g. Further, the determination unit 11f stores the history information regarding the corresponding data in the past history data 12h.

  In addition, the determination unit 11f determines whether or not the above-described additional learning is necessary based on the correlation score of the corresponding data, and when it is determined that additional learning is necessary, the determination unit 11f labels the corresponding data and performs a new classification. A class is set and stored in the additional learning data set 12i.

  Here, using FIG. 3A and FIG. 3B, a correlation score serving as an index for determining whether or not additional learning is necessary will be described. 3A and 3B are explanatory diagrams (part 1) and (part 2) of the correlation score.

  As shown in FIG. 3A, it is assumed that events # 1 and # 2 are machine-learned from an abnormal record in the past, and an abnormal classification model 12e is generated. In such a case, the abnormal classification model 12e identifies the events # 1 and # 2 in the vector space of the two-dimensional feature quantity based on the feature quantities A and B, but the unknown event # 3 is input here. In this case, as shown in the figure, such “for the unlearned event # 3, the abnormal classification model 12e outputs event # 1 or event # 2”. Assume that event # 3 is identified as event # 1.

  At this time, as shown in FIG. 3B, the abnormality classification model 12e outputs a correlation score corresponding to the distance between the sample point group of the event # 1 and the sample point of the event # 3, whereby the event # 3 of the event # 3 The degree of similarity with 1 can be estimated. The correlation score takes a value of 0 to 1. The smaller the value, the lower the similarity, and the larger the value, the higher the similarity.

  Therefore, for example, for the unknown event # 3, when the category ID “001” corresponding to the event # 1 is output by the abnormality classification model 12e, but the correlation score is extremely low (for example, 0), the determination unit 11f It can be determined that additional learning is required as a new event indicating a new abnormality. On the other hand, when the correlation score is high (for example, 0.8 or more), the determination unit 11f can determine that the additional learning is unnecessary because it is a known event.

  If the correlation score is in a gray zone that is neither extremely low nor high, the necessity of additional learning may be left to judgment based on human knowledge. In such a case, the determination unit 11f informs an external device such as a portable information terminal carried by an administrator, an operator, or the like via the notification unit 11g that there is a failure sign, the abnormality classification result, and the correlation score. The necessity of additional learning may be determined based on the response operation. For the response operation, for example, the guidance of “There is a possibility of an unknown event (correlation score: 0.5). Do you want to learn further?” Is displayed on the display unit of the external device, and “Yes” or “ “No” may be input.

  Even if the correlation score is in the gray zone, a threshold value is set in advance in the gray zone, and the determination unit 11f automatically determines whether additional learning is necessary based on the threshold value. Also good.

  Returning to the description of FIG. 2, the notification unit 11g will be described. When the notification unit 11g receives a notification of the divergence degree, the contribution rate, the category ID, and the correlation score of the corresponding data from the determination unit 11f, based on the notification and the past history data 12h, for example, the notification as illustrated in FIG. 1E Information is generated and notified to an external device.

  The update unit 11h updates the abnormality classification model 12e using the additional learning data set 12i.

  Next, a specific example of the abnormality determination result by the failure prediction apparatus 10 according to the present embodiment will be described with reference to FIGS. 4A to 4F. 4A to 4F are diagrams (No. 1) to (No. 6) illustrating specific examples of the abnormality determination result by the failure prediction apparatus 10 according to the present embodiment.

  First, as shown in FIG. 4A, during the past operation period of the failure prediction apparatus 10, “event # 1” to “event # 5” in which the deviation from the normal state exceeds a predetermined determination threshold Th1 occur. It shall have been. Then, after the abnormalities of the “event # 1” to “event # 5” are sequentially classified by the category IDs “001” to “005”, machine learning is executed as an abnormal state, and the abnormal classification model 12e is obtained. Suppose that it was generated.

  For such an abnormal classification model 12e, when the abnormal data of “event # 1” to “event # 5” are input as evaluations, the “category ID” and “correlation score” output by the abnormal classification model 12e are As shown in FIG. 4B.

  That is, as shown in FIG. 4B, the abnormality classification model 12e has a category ID “001” for “event # 1,” a category ID “002” for “event # 2,” and a “event # 3”. The category ID “003” is output, the “event # 4” is output with the category ID “004”, and the “event # 5” is output with the category ID “005”.

  In addition, the abnormality classification model 12e outputs a high value of approximately “0.8” or more for any of “event # 1” to “event # 5” for “correlation score”.

  Here, as shown in FIG. 4C, during the operation period later in time series than FIG. 4A, the degree of deviation from the normal state exceeds the determination threshold Th1, “event # 6” to “event # 10” Is assumed to occur. In addition, these “event # 6” to “event # 10” are observed in different modes as seen from the degree of deviation from the normal state, but are all classified into the same classification class (for example, category ID “003”). It was assumed that it was abnormal. Thus, in actual operation, the same abnormality is not always observed with the same degree of dissociation from the normal state.

  Abnormal data of each of such “event # 6” to “event # 10” is input as an evaluation portion to the abnormality classification model 12e of FIG. 4A, and the abnormality classification model 12e displays “category ID” and “ Assume that "correlation score" is output.

  That is, as shown in FIG. 4D, the abnormality classification model 12e outputs the category ID “003” for each of “event # 6” to “event # 10”, and the administrator is notified of this content. The operator can confirm that these are abnormalities classified by the category ID “003”.

  However, “correlation score” is a high value of “0.8” or more for “event # 6” and “event # 10”, but for “event # 7” to “event # 9”, It is a value between “0.6” and “0.8”, which is relatively low compared to “event # 6” and “event # 10”.

  Therefore, in such a case, as described above, the same abnormality is not always observed with the same degree of dissociation from the normal state, so that different anomalies have similar behavioral correlation scores. Since it may be observed, the necessity of additional learning may be determined based on human knowledge.

  That is, in the case of the example shown in FIG. 4D, at the time of notification about “event # 7” to “event # 9”, as described above, there is a possibility of an “unknown event (correlation score: 0.6 to 0.8) “Do you want to learn additional?” Is displayed, and based on the response operation from the administrator or operator who confirmed the content, the determination unit 11f determines whether additional learning is necessary or not. You may judge.

  Next, as shown in FIG. 4E, on the basis of “event # 1” to “event # 5” in FIG. 4A, first, “event # 1 and # 2 are identified by the abnormality classification model 12e that“ learned event # 1 ”. "Evaluate", then "Evaluate event # 3" with the abnormal classification model 12e that "learned additional event # 2", repeat additional learning until event # 4 in the order of ..., until "evaluate event # 5" Suppose you have gone.

  In this case, when the abnormal data of “event # 1” to “event # 5” are sequentially input as evaluations, the “category ID” and “correlation score” output by the abnormal classification model 12e are shown in FIG. It will be as shown in

  That is, as shown in FIG. 4F, since “event # 1” is already learned data, the abnormal classification model 12e has a high “correlation score” of category IDs “001” and “0.8” or higher. Is output.

  On the other hand, at the time of evaluation of each of “event # 2” to “event # 5”, since “event # 2” to “event # 5” are unknown data, respectively, the abnormal classification model 12e has a known category. Output the ID.

  At this time, if the correlation score is clearly low as in the case of “event # 2” to “event # 4”, the determination unit 11f adds “event # 2” to “event # 4” as additional learning points, respectively. Can be determined. On the other hand, as in the case of “event # 5”, the correlation score is approximately “0.6” to “0.8”, which shows a certain degree of similarity with known data (here, category ID “004”). If so, it is preferable to leave the judgment to human knowledge by displaying guidance or the like as described above.

  Next, a processing procedure executed by the failure prediction apparatus 10 will be described with reference to FIGS. 5A and 5B. FIG. 5A and FIG. 5B are flowcharts (part 1) and (part 2) showing the processing procedure executed by the failure prediction apparatus 10.

  As shown in FIG. 5A, first, the control unit 11 determines whether or not it is the first operation (step S101). Here, when it is the first operation (step S101, Yes), the control unit 11 sets a normal period (step S102). As the normal period, for example, “○ month × day to 30 days” is set on the system.

  Subsequently, the first generation unit 11c generates a normal model 12d by machine learning from the normal learning data set 12b extracted by the extraction unit 11b, that is, normal data for a normal period (step S103).

  Subsequently, the second generation unit 11d labels the abnormality learning data set 12c extracted by the extraction unit 11b, that is, past abnormality data, and sets a classification class (step S104). Then, the second generation unit 11d generates the abnormal classification model 12e from the past abnormal data in which the classification class is set by transfer learning based on the normal model 12d (step S105).

  If it is not the first operation in step S101 (step S101, No), control is transferred to step S106.

  Subsequently, as shown in FIG. 5B, the evaluation unit 11e determines the degree of divergence from the normal state based on the output value of the normal model 12d to which the evaluation data set 12f extracted by the extraction unit 11b, that is, the evaluation data is input. Is calculated (step S106).

  Then, the determination unit 11f determines whether or not the divergence calculated by the evaluation unit 11e is greater than or equal to a predetermined determination threshold value Th1 (step S107). Here, when the divergence degree is equal to or greater than the determination threshold Th1 (Yes at Step S107), the notification unit 11g notifies that there is a failure sign (Step S108).

  On the other hand, when the divergence degree is less than the determination threshold Th1 (step S107, No), the determination unit 11f determines that there is no failure sign (step S109) and ends the process.

  Further, following step S108, the evaluation unit 11e inputs the corresponding data to the abnormality classification model 12e, and derives an abnormality classification result and a correlation score (step S110). Then, the determination unit 11f determines whether or not the correlation score derived by the evaluation unit 11e is greater than or equal to a predetermined determination threshold (step S111).

  Here, when the correlation score is equal to or greater than the determination threshold value (step S111, Yes), the notification unit 11g notifies the abnormality classification result and the correlation score (step S112), and the process is terminated.

  On the other hand, when the correlation score is less than the determination threshold (No at Step S111), the determination unit 11f determines whether additional learning is necessary (Step S113). Here, when it is determined that additional learning is necessary (step S113, Yes), the determination unit 11f labels the corresponding data and adds a new classification class (step S114).

  Then, the update unit 11h updates the abnormal classification model 12e from the corresponding data stored in the additional learning data set 12i by the determination unit 11f (step S115), and the process ends. If it is determined that no additional learning is required (No at step S113), the process is terminated.

  By the way, the failure prediction apparatus 10 according to the embodiment described above is realized by a computer 60 having a configuration as shown in FIG. 6, for example. FIG. 6 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the failure prediction apparatus 10. The computer 60 includes a central processing unit (CPU) 61, a random access memory (RAM) 62, a read only memory (ROM) 63, a hard disk drive (HDD) 64, a communication interface (I / F) 65, an input / output interface (I). / F) 66 and a media interface (I / F) 67.

  The CPU 61 operates based on a program stored in the ROM 63 or the HDD 64 and controls each unit. The ROM 63 stores a boot program executed by the CPU 61 when the computer 60 is activated, a program depending on the hardware of the computer 60, and the like.

  The HDD 64 stores a program executed by the CPU 61, data used by the program, and the like. The communication interface 65 receives data from other devices via the communication network and sends the data to the CPU 61, and transmits the data generated by the CPU 61 to other devices via the communication network.

  The CPU 61 controls an output device such as a display and a printer and an input device such as a keyboard and a mouse via the input / output interface 66. The CPU 61 acquires data from the input device via the input / output interface 66. Further, the CPU 61 outputs the generated data to the output device via the input / output interface 66.

  The media interface 67 reads a program or data stored in the recording medium 68 and provides it to the CPU 61 via the RAM 62. The CPU 61 loads the program from the recording medium 68 onto the RAM 62 via the media interface 67 and executes the loaded program. The recording medium 68 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or PD (Phase change rewritable disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. Etc.

  For example, when the computer 60 functions as the failure prediction apparatus 10 according to the embodiment, the CPU 61 of the computer 60 implements each function of the control unit 11 by executing a program loaded on the RAM 62. The HDD 64 stores data in the storage unit 12. The CPU 61 of the computer 60 reads these programs from the recording medium 68 and executes them, but as another example, these programs may be acquired from other devices via a communication network.

  As described above, the failure prediction apparatus 10 according to the embodiment includes the extraction unit 11b, the first generation unit 11c, the second generation unit 11d, the evaluation unit 11e, the determination unit 11f, and the notification unit 11g. Prepare.

  The extraction unit 11b includes, for sensor data of a plurality of sensors S-1 to Sn provided in the target machine 100 (corresponding to an example of “mechanical equipment”), a normal state during normal operation of the target machine 100, An abnormal state component at the time of occurrence of an abnormality and an evaluation component at an arbitrary evaluation time are extracted.

  The 1st production | generation part 11c produces | generates the normal model 12d which modeled the correlation between the sensors S-1 to Sn at the time of normal operation by performing the machine learning using the part for a normal state. The second generation unit 11d generates an abnormal classification model 12e that models an abnormal pattern when an abnormality occurs by executing transfer learning using an abnormal state based on the normal model 12d.

  The evaluation unit 11e evaluates the degree of deviation from the normal state of the target machine 100 based on the output value of the normal model 12d obtained by inputting the evaluation amount to the normal model 12d.

  If the determination unit 11f determines that there is a failure sign in the target machine 100 based on the degree of deviation, the determination unit 11f performs an abnormality based on the output value of the abnormality classification model 12e obtained by inputting the evaluation amount to the abnormality classification model 12e. Determine the pattern. The notification unit 11g notifies the determination result of the determination unit 11f including the abnormal pattern.

  Therefore, according to the failure prediction apparatus 10 according to the present embodiment, the failure prediction accuracy of the target machine 100 can be improved.

  By the way, in the above-described embodiment, the case where at least an external device is notified when it is determined that there is a failure sign is taken as an example, but the present invention is not limited thereto. For example, the determination unit 11f may determine whether or not notification is necessary depending on whether or not it is important among those determined to have a failure sign. Whether or not it is important may be distinguished by, for example, a category ID indicating an abnormal pattern. In addition, a discriminant model for discriminating insignificant abnormal patterns is generated by machine learning in advance, and an evaluation portion determined to have a sign of failure is input to the discriminant model. It may be obtained whether the pattern is an abnormal pattern / unimportant abnormal pattern that does not require notification.

  In the above-described embodiment, deep learning is used as a machine learning algorithm, but the algorithm to be used is not limited. Therefore, machine learning may be executed by a regression analysis method such as support vector regression using a pattern classifier such as SVM (Support Vector Machine) to generate the normal model 12d and the abnormal classification model 12e. Here, the pattern discriminator is not limited to SVM, and may be, for example, AdaBoost. A random forest or the like may be used.

  Further effects and modifications can be easily derived by those skilled in the art. Thus, the broader aspects of the present invention are not limited to the specific details and representative embodiments shown and described above. Accordingly, various modifications can be made without departing from the spirit or scope of the general inventive concept as defined by the appended claims and their equivalents.

DESCRIPTION OF SYMBOLS 1 Failure prediction system 10 Failure prediction apparatus 11 Control part 11a Collection part 11b Extraction part 11c 1st generation part 11d 2nd generation part 11e Evaluation part 11f Determination part 11g Notification part 11h Update part 12 Storage part 12a Collected data 12b Normal learning data Set 12c Abnormal learning data set 12d Normal model 12e Abnormal classification model 12f Evaluation data set 12g Evaluation information 12h Past history data 12i Additional learning data set 100 Target machine S-1 to Sn sensor

Claims (8)

An extraction step for extracting a normal state component during normal operation of the mechanical facility, an abnormal state component at the occurrence of an abnormality, and an evaluation component at an arbitrary evaluation time from sensor data of a plurality of sensors provided in the mechanical facility; ,
A first generation step of generating a normal model that models the correlation between the sensors during the normal operation by performing machine learning using the normal state; and
A second generation step of generating an abnormal classification model that models an abnormal pattern at the time of occurrence of the abnormality by executing transfer learning using the abnormal state component based on the normal model;
An evaluation step for evaluating the degree of deviation from the normal state of the mechanical equipment based on the output value of the normal model obtained by inputting the evaluation portion to the normal model;
When it is determined that there is a failure sign in the mechanical equipment based on the degree of deviation, the abnormality pattern is determined based on an output value of the abnormality classification model obtained by inputting the evaluation portion to the abnormality classification model A determination step to
A failure prediction method comprising: a notification step of notifying a determination result of the determination step including the abnormal pattern. An update step of updating the abnormality classification model by executing machine learning using the evaluation portion estimated to indicate a new abnormality pattern based on an output value of the abnormality classification model as an additional learning portion The failure prediction method according to claim 1. The second generation step includes
When the evaluation portion is input, the abnormality classification model is generated so that an identifier of the abnormal pattern and a similarity degree of the evaluation portion with respect to the abnormal pattern are output,
The determination step includes
The failure prediction method according to claim 2, wherein the notifying step notifies the determination result including the similarity. The determination step includes
4. The abnormality classification model is updated in the update step, when the similarity is less than a predetermined determination threshold, with the evaluation corresponding to the similarity as the additional learning. The failure prediction method described. The determination step includes
The update process is configured to notify the notification process so that a response operation as to whether or not to update can be accepted from the outside before the update process updates the abnormality classification model by the additional learning. 4. The failure prediction method according to 4. The determination step includes
The failure prediction method according to claim 1, wherein the necessity of notification by the notification step is determined according to the abnormality pattern. An extraction unit that extracts, from sensor data of a plurality of sensors provided in a mechanical facility, a normal state portion during normal operation of the mechanical facility, an abnormal state portion when an abnormality occurs, and an evaluation portion at an arbitrary evaluation time ,
A first generation unit that generates a normal model that models the correlation between the sensors during the normal operation by executing machine learning using the normal state;
A second generation unit that generates an abnormality classification model that models an abnormal pattern at the time of occurrence of the abnormality by performing transfer learning using the abnormal state component based on the normal model;
An evaluation unit that evaluates the degree of deviation from the normal state of the mechanical equipment based on the output value of the normal model obtained by inputting the evaluation part to the normal model;
When it is determined that there is a failure sign in the mechanical equipment based on the degree of deviation, the abnormality pattern is determined based on an output value of the abnormality classification model obtained by inputting the evaluation portion to the abnormality classification model A determination unit to perform,
A failure prediction device comprising: a notification unit that notifies a determination result of the determination unit including the abnormal pattern. An extraction procedure for extracting a normal state component during normal operation of the mechanical facility, an abnormal state component at the time of occurrence of an abnormality, and an evaluation component at an arbitrary evaluation time from sensor data of a plurality of sensors provided in the mechanical facility; ,
A first generation procedure for generating a normal model that models the correlation between the sensors during the normal operation by performing machine learning using the normal state;
A second generation procedure for generating an abnormal classification model that models an abnormal pattern at the time of occurrence of the abnormality by performing transfer learning using the abnormal state component based on the normal model;
An evaluation procedure for evaluating the degree of deviation from the normal state of the mechanical equipment based on the output value of the normal model obtained by inputting the evaluation portion to the normal model;
When it is determined that there is a failure sign in the mechanical equipment based on the degree of deviation, the abnormality pattern is determined based on an output value of the abnormality classification model obtained by inputting the evaluation portion to the abnormality classification model Judgment procedure to
A failure prediction program that causes a computer to execute a notification procedure for reporting a determination result of the determination procedure including the abnormal pattern.

Images