JP2020052459A - Failure prediction method, failure prediction device and failure prediction program - Google Patents

Abstract

To improve the failure prediction accuracy of a machinery facility.SOLUTION: A failure prediction method according to the embodiment includes an extraction step, a generation step, an evaluation step, and a determination step. In the extraction step, a normal state part during a normal operation of a machinery facility among sensor data of a plurality of sensors provided in the machinery facility having a plurality of operation modes, and an evaluation part at an arbitrary evaluation time are extracted for each operation mode. In the generation step, multiple normal models that model the correlation between the sensors corresponding to the respective operation modes are generated by executing machine learning using each of the normal state part. In the evaluation step, the degree of divergence from the normal state of the machinery facility is calculated for each operation mode on the basis of an output value of the normal model obtained by inputting each of the evaluation part into the corresponding normal model. In the determination step, a failure sign of the machinery facility is determined based on an integration result integrating the degree of divergence for each operation mode.SELECTED DRAWING: Figure 2

Description

  The disclosed embodiments relate to a failure prediction method, a failure prediction device, and a failure prediction program.

  2. Description of the Related Art Conventionally, there has been known a technology for predicting occurrence of a failure by monitoring a sensor value of a sensor provided in the mechanical facility to detect a failure sign (for example, see Patent Document 1).

  Conventionally, for machine equipment, rule information including a condition for performing an intervention operation is generated by executing machine learning based on past intervention operation results, and the rule information 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 above-described rule information corresponds to a so-called prediction model.

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

JP 2011-230634 A JP 2017-049801 A

  However, the above-described conventional technology has room for further improvement in improving the accuracy of predicting the failure of mechanical equipment.

  Specifically, it is practically difficult to collect and cover machine learning data indicating the status of all failures at the initial stage of generating a prediction model, such as the initial stage of operation of mechanical equipment. In particular, when the mechanical equipment is a large mechatronics machine such as a large refrigerator or a plant, the number of sensors provided is enormous, which is more difficult.

  Therefore, instead of learning the data at the time of failure, a normal model that learns the data at normal time and models the normal state of the machinery and equipment so that failure prediction is possible even with machine learning with a small amount of data is possible. It is conceivable to generate and use this.

  However, the mechanical equipment may have a plurality of operation modes, and even in a normal state, data tend to be different depending on the operation mode in many cases. For this reason, even if a normal model is used, if the machine equipment has a plurality of operation modes, machine learning with a small amount of data may not be able to capture the characteristics of the normal state, and the accuracy of the normal model may not be secured. There is.

  An aspect of the embodiment is made in view of the above, and an object of the present invention is to provide a failure prediction method, a failure prediction device, and a failure prediction program that can improve the failure prediction accuracy of mechanical equipment.

  A failure prediction method according to one aspect of an embodiment includes an extraction step, a generation step, an evaluation step, and a determination step. The extraction step, of the sensor data of a plurality of sensors provided in the mechanical equipment having a plurality of operation modes, the normal state during normal operation of the mechanical equipment, and the evaluation at the time of any evaluation, the said Extract for each operation mode. The generation step generates a plurality of normal models that model the correlation between the sensors corresponding to each of the operation modes by executing machine learning using each of the normal states. The evaluation step calculates a degree of deviation from a normal state of the mechanical equipment for each of the operation modes based on an output value of the normal model obtained by inputting each of the evaluation components to the corresponding normal model. I do. The determining step determines a failure sign of the mechanical equipment based on an integrated result obtained by integrating the divergence degrees for each of the operation modes.

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

FIG. 1A is a schematic explanatory diagram (part 1) of a 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. 3 is a diagram illustrating an example of the mode separation information. FIG. 4 is an explanatory diagram of a determination process performed by the determination unit. FIG. 5A is a diagram illustrating an example of selecting a degree of deviation. FIG. 5B is a diagram (part 1) illustrating the operation of the extraction unit when the operation mode is switched. FIG. 5C is a diagram (part 2) illustrating the operation of the extraction unit when the operation mode is switched. FIG. 5D is an explanatory diagram of the operation of the determination unit when the operation mode is switched. FIG. 6A is a flowchart (part 1) illustrating a processing procedure executed by the failure prediction device according to the embodiment. FIG. 6B is a flowchart (part 2) illustrating a processing procedure executed by the failure prediction device according to the embodiment. FIG. 7 is a hardware configuration diagram illustrating an example of a computer that implements the functions of the failure prediction device according to the embodiment.

  Hereinafter, with reference to the accompanying drawings, 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. The present invention is not limited by the embodiments described below.

  In the following, the mechanical equipment to be subjected to the failure sign determination is referred to as “target machine 100”.

  First, an outline of a failure prediction method according to the embodiment will be described with reference to FIGS. 1A to 1E. 1A to 1E are schematic explanatory diagrams (part 1) to (part 5) of a failure prediction method according to the embodiment.

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

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

  Here, the “normal state portion” refers to a sensor data group in a predetermined period (hereinafter, sometimes referred to as a “normal period”) in which the target machine 100 is in a normal state in an initial operation stage 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.

  Then, in the failure prediction method according to the embodiment, as shown in FIG. 1B, sensor data of each of the sensors S-1 to Sn indicating the correlation of “evaluation” is input to the normal model 12d ′. Then, the correlation deviation amount 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 (hereinafter, may be simply referred to as “degree of deviation”) is evaluated based on the amount of deviation of the correlation (step S2).

  If the degree of deviation is large, it is possible to predict the failure of the target machine 100 by indicating a failure sign. Here, the “evaluation portion” is, for example, a predetermined period to be evaluated (hereinafter, sometimes referred to as an “evaluation period”) in a sensor data group output from the target machine 100 and accumulated during operation. And a sensor data group output from the target machine 100 in real time.

  As described above, in the failure prediction method according to the 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 in the normal state. It can be obtained from the output value of the normal model 12d '. Then, a failure of the target machine 100 is predicted based on the magnitude of the deviation based on the output value. For example, if the degree of divergence is equal to or greater than a predetermined determination threshold value indicating a sign of failure, an abnormality is notified to the user.

  Therefore, according to the failure prediction method according to the embodiment, for example, a failure prediction method using a prediction model generated by machine learning based on sensor data at the time of occurrence of a failure becomes necessary. Complicated steps such as implementation of modeling are not required. That is, according to the failure prediction method according to the embodiment, the failure sign of the target machine 100 can be easily grasped.

  In the failure prediction method according to the embodiment, as shown in FIG. 1A, machine learning including time variation of each sensor data, that is, time-series correlation in a feature vector (hereinafter, simply referred to as “vector”) is performed. By doing so, it is possible to grasp a sign of a failure that appears in the time variation of the sensor data. As a result, a sign of failure of the target machine 100 can be accurately grasped.

  Incidentally, the target machine 100 may have a plurality of operation modes. In such a case, as shown in FIG. 1C, the target machine 100 operates while the operation mode is “mode-switched” over time. Here, an example is shown in which the target machine 100 is operating while switching between the two modes of the “first operation mode” and the “second operation mode”, but of course, there are three or more operation modes. You may. The operation mode is, for example, a heating operation or a cooling operation in a cooling / heating machine, for example.

  Even if the target machine 100 is in a normal state in which the target machine 100 is operating normally without failure, if such an operation mode is different, as shown in FIG. 1D, a certain one of the sensors S-1 to Sn is used. (Sensors S-1 to S-3 in the drawing) may have different tendencies in sensor data.

  That is, even in the same normal state, if the operation mode is different, the correlation between the sensors S-1 to Sn may be different. The difference in the characteristics of the normal state due to the difference in the operation modes can be captured if the data set for machine learning that generates the normal model 12d 'is rich enough to completely cover all the normal states in all the operation modes. There is sex.

  However, it is difficult to collect such a sufficient data set in advance in the initial stage of operation of the target machine 100 or the like. Therefore, the normal model 12d 'is generated using only a data set that can be collected in advance, but in such a case, the normal model 12d' cannot capture the difference in the characteristics of the normal state due to the difference in the operation mode described above. There is a risk. This is because the normal state of the target machine 100 is to be modeled with one normal model 12d 'regardless of the difference in the operation mode.

  Therefore, in the failure prediction method according to the embodiment, a plurality of normal models 12d-1, 12d-2,... Corresponding to the respective operation modes of the target machine 100 are generated. Then, a failure sign of the target machine 100 is determined while integrating the divergence obtained by inputting the data set to be evaluated into the corresponding normal model.

  Specifically, as shown in FIG. 1E, in the failure prediction method according to the embodiment, first, a normal model is obtained by machine learning using each data set of “normal state” in each operation mode in the “normal period”. Generate the group 12d.

  The normal model group 12d includes, for example, a first normal model 12d-1 generated by machine learning using “normal state components” in the “first operation mode” and a “normal state” in the “second operation mode”. And the second normal model 12d-2 generated by machine learning using "minute".

  Then, in the failure prediction method according to the embodiment, each data set of “evaluation amount” in each operation mode in the “evaluation period” is stored in the corresponding normal models 12d-1, 12d-2,. Enter For example, the “evaluation part” in the “first operation mode” is input to the first normal model 12d-1, and the “deviation” with respect to the input is output. In addition, for example, the “evaluation part” in the “second operation mode” is input to the second normal model 12d-2, and the “deviation degree” with respect to the input is output. As described above, in the failure prediction method according to the embodiment, “calculation degree is calculated for each mode” as shown in the figure.

  Then, for example, by mapping the divergence calculated for each mode on the same time axis, the divergence in the entire normal model group 12d is integrated, and it is determined whether or not there is a sign of failure. A specific example of such determination processing will be described later with reference to FIG.

  Thus, according to the failure prediction method according to the embodiment, even when the target machine 100 has a plurality of operation modes, the failure prediction accuracy of the target machine 100 can be improved.

  Further, in the failure prediction method according to the embodiment, for example, when a plurality of divergence degrees due to a difference in operation mode are calculated in a predetermined period in which one divergence degree should be derived, one divergence degree is calculated according to a predetermined rule. Choose a degree. Such a specific example will be described later with reference to FIG. 5A.

  In addition, for example, immediately after “mode switching” is performed, the sensor data may show an unstable tendency due to the influence of the switching. In such a case, the sensor data which is in a normal state but indicates an outlier is likely to be included in the “normal state” for machine learning and the “evaluation part” to be evaluated.

  In preparation for such a case, in the failure prediction method according to the embodiment, the sensor data immediately after the "mode switching" is handled according to a predetermined rule, and the failure prediction accuracy is ensured. For example, in the failure prediction method according to the embodiment, a predetermined period immediately after “mode switching” is treated as a so-called dead zone. Specific examples of such handling will be described later with reference to FIGS. 5B to 5D.

  Hereinafter, the configuration of the failure prediction system 1 according to the embodiment 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 embodiment. In FIG. 2, components necessary for explaining the features of the present embodiment are represented by functional blocks, and descriptions of general components are omitted.

  In other words, each component illustrated in FIG. 2 is a functional concept 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 illustrated one, and all or a part of the functional blocks may be functionally or physically distributed in arbitrary units according to various loads and usage conditions. -It is possible to integrate and configure.

  In the description using FIG. 2, the description of the components already described above may be simplified or omitted.

  As illustrated in FIG. 2, the failure prediction system 1 includes a failure prediction device 10 and a target machine 100. The failure prediction device 10 and the target machine 100 are connected to a network and provided so as to be communicable by wire or wirelessly. The failure prediction device 10 is provided so that sensor data from the target machine 100 can be appropriately collected.

  The failure prediction device 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 generation unit 11c, an evaluation unit 11d, a determination unit 11e, and a notification unit 11f.

  The storage unit 12 is a storage device such as a hard disk drive, a data flash, a non-volatile memory, and a register. The storage unit 12 includes collected data 12a, mode separation information 12b, a learning data set group 12c, a normal model group 12d, The data set group 12e and the evaluation information 12f are stored.

  The control unit 11 performs overall control of the failure prediction device 10. The collection unit 11a collects sensor data from the sensor group of the target machine 100 at a predetermined cycle and stores the collected data in the collected 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 aging or the like.

  The extraction unit 11b extracts a data set for a normal state for each operation mode from the collected data 12a based on the normal period and mode separation information 12b set at the first operation, and stores the data set in the learning data set group 12c.

  Further, the extracting unit 11b extracts a data set for each operation mode from the collected data 12a based on the evaluation period and the mode separation information 12b specified by the user, for example, and stores the data set in the evaluation data set group 12e.

  The mode separation information 12b is information relating to separation of the sensor data group stored in the collected data 12a according to the operation mode. Here, an example of the mode separation information 12b will be described with reference to FIG.

  FIG. 3 is a diagram illustrating an example of the mode separation information 12b. As shown in FIG. 3, the mode separation information 12b is information in which a combination of sensors for separating each operation mode and a threshold value thereof are associated.

  In the example of FIG. 3, for example, a combination of sensors including at least the sensors S-1 and S-2 is associated with the first operation mode, and the sensor data of the sensor S-1 exceeds 30 (see FIG. 3). The extraction unit 11b extracts, as a data set in the first operation mode, a sensor data group in which the sensor data of other sensors in the combination also satisfies the threshold condition.

  Similarly, in the example of FIG. 3, for example, a combination of sensors including at least the sensor S-1 is associated with the second operation mode, and the sensor data of the sensor S-1 is 30 or less (in the figure, “ ≦ 30 ”), the extraction unit 11b extracts a sensor data group in which the sensor data of the other sensors in the combination also satisfies the threshold condition.

  Note that, here, “combination” is described, but the operation mode may be separated based on the sensor data of only one sensor.

  Returning to the description of FIG. 2, the generation unit 11c will be described. The generation unit 11c executes machine learning using the learning data set group 12c to generate a normal model group 12d. That is, the generation unit 11c generates the first normal model 12d-1 from the learning data set indicating the normal state in the first operation mode, for example. Further, the generation unit 11c generates the second normal model 12d-2, for example, from a learning data set indicating a normal state component in the second operation mode (see FIG. 1E). As described above, the generation unit 11c executes machine learning for each operation mode, and generates normal models 12d-1 to 12d-n corresponding to each operation mode.

  The evaluation unit 11d inputs the data set for the evaluation for each operation mode extracted by the extraction unit 11b included in the evaluation data set group 12e to the corresponding normal models 12d-1 to 12d-n, respectively. Receive the output result of Then, the evaluation unit 11d calculates various evaluation values for determining a failure sign based on the received output result, and stores the calculated evaluation values in the evaluation information 12f. The evaluation value includes the degree of deviation from the normal state for each operation mode.

  Specifically, the evaluation unit 11d determines an error between the input and the output when the data set for the evaluation for each operation mode is input to the corresponding normal models 12d-1 to 12d-n, that is, based on the deviation amount of the correlation. Then, the degree of deviation for each operation mode is calculated and stored in the evaluation information 12f.

  The determination unit 11e refers to the evaluation information 12f, maps the degree of deviation for each operation mode on, for example, the same time axis, integrates the degree of deviation for the entire normal model group 12d during the evaluation period, and has a sign of failure. It is determined whether or not.

  Here, an example of the determination process performed by the determination unit 11e will be described with reference to FIG. FIG. 4 is an explanatory diagram of a determination process performed by the determination unit 11e. The upper part of FIG. 4 shows, as a comparative example, a case where the normal state of the target machine 100 is modeled by one normal model 12d '(see FIG. 1B) regardless of the difference in the operation mode. The lower part of FIG. 4 shows the case of the embodiment.

  As shown in FIG. 4, the determination unit 11e maps, for example, each of the calculated degrees of deviation from the normal state on the same time axis. Then, a point in time exceeding a predetermined determination threshold value Th1 is determined to be a failure sign.

  Here, in the case of the comparative example in which the degree of divergence is calculated by one normal model 12d 'without considering the difference in the operation mode, the difference in the characteristic of the normal state due to the difference in the operation mode can be sufficiently learned with a small amount of data. May not be. Then, when the target machine 100 is operated while switching the operation mode, for example, as shown in the upper part of FIG. 4, the degree of deviation from the normal state is constant along the time axis despite the normal state. It may show an unstable tendency to scatter.

  In this case, the time point exceeding the determination threshold value Th1 is present at every location, and even if the determination unit 11e notifies the user of the determination result, it is difficult for the user to identify the location where the failure sign has truly appeared. .

  On the other hand, in the failure prediction method according to the embodiment, the degree of divergence is calculated for each operation mode by the normal models 12d-1 to 12dn for each operation mode in consideration of the difference between the operation modes. It is possible to sufficiently learn the difference in the characteristics of the normal state due to the difference even with a small amount of data.

  Then, as shown in the lower part of FIG. 4, even when the target machine 100 is operated while switching the operation mode, the deviation degree from the normal state can be calculated as a stable value compared to the comparative example. it can. Then, the determination unit 11e notifies the user of, for example, the mapping result of the divergence degree and the time point at which the determination threshold value Th1 is exceeded, so that the user sets the M1, M2, and M3 portions in the drawing as the locations where the failure sign has occurred. Can be easily identified.

  Returning to the description of FIG. 2, the notification unit 11f will be described. When the notification of the determination result is received from the determination unit 11e, the notification unit 11f generates notification information including a mapping result of the divergence degree and a time point exceeding the determination threshold Th1 as shown in the lower part of FIG. Notify the device.

  By the way, in the actual operation of the target machine 100, there are cases where it is sufficient that the user receives a notification once a day whether or not there is a failure sign. In such a case, for example, if switching of the operation mode occurs once or more in one day, in the embodiment, the divergence is calculated for each operation mode. You have to choose.

  FIG. 5A is a diagram illustrating an example of selecting a degree of deviation. As shown in FIG. 5A, it is assumed that the target machine 100 operates in the first operation mode in the morning and operates in the second operation mode in the afternoon, for example, on a certain day.

  Then, the evaluation unit 11d calculates “0.1” as the divergence corresponding to the first operation mode in the morning, and calculates “0.7” as the divergence corresponding to the second operation mode in the afternoon. I do.

  In such a case, if the notification of whether or not there is a sign of failure only needs to be made once a day, the determination unit 11e adopts the larger divergence as one day and determines whether the difference exceeds the determination threshold Th1. Is determined, and the result is notified to the user.

  That is, in such a case, by notifying the user of the determination result based on the degree of divergence of a value as large as possible, it is possible to prompt the user to take an action immediately when a failure is actually predicted.

  As described above, for example, immediately after “mode switching” is performed, there is a case where sensor data tends to be unstable due to the influence of the switching. In such a case, the sensor data which is in a normal state but indicates an outlier is likely to be included in the “normal state” for machine learning and the “evaluation part” to be evaluated.

  In preparation for such a case, the failure prediction device 10 according to the embodiment handles the sensor data immediately after the “mode switching” or the like according to a predetermined rule, and secures the failure prediction accuracy. This will be specifically described with reference to FIGS. 5B and 5C.

  FIGS. 5B and 5C are explanatory diagrams (part 1) and (part 2) of the operation of the extraction unit 11b when the operation mode is switched. As shown in FIG. 5B, it is assumed that “mode switching” from “first operation mode” to “second operation mode” has occurred at time t1.

  In such a case, immediately after switching to the “second operation mode” in which the sensor data tends to be unstable, for example, the extraction unit 11b sets the predetermined period α from the time t1 as the “dead zone”.

  Here, the term “dead zone” indicates, for example, that the extraction unit 11b does not extract the sensor data during the predetermined period α for learning or evaluation. That is, by not learning or evaluating the predetermined period α in which the sensor data is likely to be unstable, for example, it is possible to prevent an outlier from being included in the sensor data group indicating a normal state, and to improve the failure prediction accuracy. Can be secured. Note that at least the accuracy of the normal model group 12d may be ensured by not extracting at least for learning.

  Further, instead of not extracting the sensor data at all during the predetermined period α, only a part of the sensor data may not be extracted. For example, for each of the sensor data during the predetermined period α, the rate of change (or error) before “mode switching” is calculated, and a normal distribution of the rate of change is taken as shown in FIG. The sensor data corresponding to the standard deviation 2σ or more in the distribution (see the M4 part in the figure) may not be extracted.

  Thus, a steep change is exhibited in the predetermined period α, and learning or evaluation is not performed on sensor data that is estimated to be unstable, thereby contributing to securing accuracy of failure prediction. .

  Although the standard deviation is equal to or larger than 2σ in the example of FIG. 5C, this is merely an example, and does not limit the criterion for determining not to extract the sensor data.

  5B and 5C show an operation example of the extraction unit 11b for the predetermined period α, but the determination unit 11e may perform a predetermined operation for the predetermined period α. FIG. 5D is an explanatory diagram of the operation of the determination unit 11e when the operation mode is switched.

  For example, as shown in FIG. 5D, the determination unit 11e determines the degree of deviation from the normal state corresponding to the predetermined period α from the time “t1” of “mode switching” by a determination threshold for determining that there is a sign of failure. Th1 may be increased.

  As described above, the failure prediction accuracy can also be improved by increasing the determination threshold Th1 to reduce the sensitivity for determining the failure sign for the deviation calculated based on the sensor data that is likely to be unstable in the predetermined period α. Can help secure it.

  Next, a processing procedure executed by the failure prediction device 10 according to the embodiment will be described with reference to FIGS. 6A and 6B. 6A and 6B are a flowchart (part 1) and (part 2) illustrating a processing procedure executed by the failure prediction device 10 according to the embodiment.

  As shown in FIG. 6A, first, the control unit 11 determines whether or not it is the first operation (step S101). Here, if it is the first operation (step S101, Yes), the control unit 11 sets a normal period (step S102). The normal period is set in the system, for example, “30 days from month x day”. If it is not the first operation (step S101, No), the control is transferred to step S105 (see FIG. 6B). Step S105 and subsequent steps will be described later.

  Then, the extraction unit 11b extracts a learning data set for each mode in the normal period based on the mode separation information 12b.

  Subsequently, the control unit 11 executes a loop process of step S104. Step S104 is repeated for the number of data sets for each mode extracted in step S103. In step S1041, the generation unit 11c generates (any one of) the normal models 12d-1 to 12dn by machine learning. Then, a mode is associated with the generated normal models 12d-1 to 12d-n (step S1042).

  When the loop processing of step S104 ends, step S105 is subsequently executed. As shown in FIG. 6B, in step S105, the extraction unit 11b extracts an evaluation data set for each mode in the evaluation period based on the mode separation information 12b.

  Subsequently, the control unit 11 executes a loop process of step S106. Step S106 is repeated for the number of data sets for each mode extracted in step S105. In step S1061, the evaluation unit 11d calculates the degree of divergence using (one of) the normal models 12d-1 to 12dn corresponding to the mode corresponding to the evaluation data set (step S1061).

  When the loop process of step S106 ends, the determination unit 11e determines a failure sign based on the calculated degree of divergence for each mode, notifies the determination result (step S107), and ends the process.

  By the way, the failure prediction device 10 according to the above-described embodiment is realized by, for example, a computer 60 having a configuration as shown in FIG. FIG. 7 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the failure prediction device 10 according to the embodiment. The computer 60 includes a CPU (Central Processing Unit) 61, a RAM (Random Access Memory) 62, a ROM (Read Only Memory) 63, a HDD (Hard Disk Drive) 64, a communication interface (I / F) 65, and 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 started, a program depending on 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 another device via the communication network, sends the data to the CPU 61, and transmits the data generated by the CPU 61 to the other device 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 an 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 the program or data 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), a 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. And so on.

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

  As described above, the failure prediction device 10 according to the embodiment includes the extraction unit 11b, the generation unit 11c, the evaluation unit 11d, and the determination unit 11e.

  The extraction unit 11b performs a normal operation of the target machine 100 among the sensor data of the plurality of sensors S-1 to Sn provided in the target machine 100 having a plurality of operation modes (corresponding to an example of “machine equipment”). The normal state component at the time and the evaluation component at an arbitrary evaluation are extracted for each operation mode.

  The generation unit 11c executes machine learning using each of the normal states, thereby normalizing models 12d-1 to 12d that model the correlation between the sensors S-1 to Sn corresponding to the respective operation modes. Generate a plurality of −n.

  The evaluation unit 11d determines the normal state of the target machine 100 based on the output values of the normal models 12d-1 to 12d-n obtained by inputting the respective evaluations to the corresponding normal models 12d-1 to 12d-n. Is calculated for each operation mode.

  The determining unit 11e determines a sign of failure of the target machine 100 based on an integration result obtained by integrating the degrees of deviation for each operation mode.

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

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

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

Reference Signs List 1 failure prediction system 10 failure prediction device 11 control unit 11a collection unit 11b extraction unit 11c generation unit 11d evaluation unit 11e determination unit 11f notification unit 12 storage unit 12a collected data 12b mode separation information 12c learning data set group 12d normal model group 12e Evaluation data set group 12f Evaluation information 100 Target machine S-1 to Sn Sensor

Claims (5)

Of the sensor data of the plurality of sensors provided in the mechanical equipment having a plurality of operation modes, a normal state part during a normal operation of the mechanical equipment, and an evaluation part at an arbitrary evaluation are extracted for each of the operation modes. Extraction process,
A generation step of generating a plurality of normal models that model the correlation between the sensors corresponding to the respective operation modes by executing machine learning using each of the normal states,
An evaluation step of calculating a degree of deviation from a normal state of the mechanical equipment for each of the operation modes, based on an output value of the normal model obtained by inputting each of the evaluation components to the corresponding normal model,
A determination step of determining a sign of failure of the mechanical equipment based on an integrated result obtained by integrating the degrees of deviation for each of the operation modes. The extracting step includes:
When the operation mode is changed from one operation mode to another operation mode other than the one operation mode, all or a part of the sensor data for a predetermined period from the time of the change is converted into at least the normal state. The failure prediction method according to claim 1, wherein the failure prediction is not performed. The determining step includes:
When the deviation exceeds a predetermined determination threshold, it is determined that there is a sign of failure in the mechanical equipment, and the operation mode is switched from one operation mode to another operation mode other than the one operation mode. 3. The failure prediction method according to claim 1, wherein in the case of the failure, the degree of divergence corresponding to a predetermined period from the switching time is determined by increasing the determination threshold. 4. Of the sensor data of the plurality of sensors provided in the mechanical equipment having a plurality of operation modes, a normal state part during a normal operation of the mechanical equipment, and an evaluation part at an arbitrary evaluation are extracted for each of the operation modes. An extraction unit,
A generation unit that generates a plurality of normal models that model the correlation between the sensors corresponding to the operation modes by executing machine learning using each of the normal states.
An evaluation unit that calculates a degree of deviation from a normal state of the mechanical equipment for each of the operation modes based on an output value of the normal model obtained by inputting each of the evaluation components to the corresponding normal model. ,
A failure prediction device comprising: a determination unit configured to determine a failure sign of the mechanical equipment based on an integration result obtained by integrating the degrees of deviation for each of the operation modes. Of the sensor data of the plurality of sensors provided in the mechanical equipment having a plurality of operation modes, a normal state part during a normal operation of the mechanical equipment, and an evaluation part at an arbitrary evaluation are extracted for each of the operation modes. Extraction procedure,
A generation procedure of generating a plurality of normal models that model the correlation between the sensors corresponding to the operation modes by executing machine learning using each of the normal states,
An evaluation procedure of calculating a degree of deviation from a normal state of the mechanical equipment for each of the operation modes, based on an output value of the normal model obtained by inputting each of the evaluations to the corresponding normal model; and ,
A failure prediction program characterized by causing a computer to execute a determination procedure of determining a failure sign of the mechanical equipment based on an integration result obtained by integrating the degrees of deviation for each of the operation modes.