JP2018005393A - Failure sign determination method, failure sign determination device and failure sign determination program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To easily and accurately capture deviation from a normal state indicating a sign of failure of mechanical equipment.SOLUTION: A failure sign determination method includes a collection process, an extraction process, a setting process, a generation process, a calculation process, and a determination process. The extraction process extracts, from data sets, data sets for a predetermined normal period during which mechanical equipment was in a normal state and data sets for a predetermined latest period in the past. After associating first additional information with the data sets for the normal period and unconditionally associating second additional information indicating failure with the data sets for the latest period, the setting process sets two groups in which data sets for the normal period coexist with data sets for the latest period. The generation process generates, through learning using one of the two groups, a prediction model for predicting additional information corresponding input data. The calculation process calculates a probability of being predicted that additional information in a prediction result of a prediction model obtained by using the other of the two groups as input data is the second additional information. The determination process determines a sign of failure of the mechanical equipment based on the probability.SELECTED DRAWING: Figure 2A

Description

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

  2. Description of the Related Art Conventionally, a technique for detecting a failure sign by monitoring a sensor value of a sensor provided in a mechanical facility is known (for example, see Patent Document 1).

  The technique disclosed in Patent Document 1 compares a sensor value of a sensor mounted on a vehicle with a normal threshold value, and determines whether or not an abnormality has occurred in the sensor based on the comparison result. Then, when it is determined that an abnormality has occurred in the sensor, the evaluation index for evaluating the failure sign of the sensor is calculated using the difference between the abnormality occurrence duration and the sensor value and the normal threshold value, A failure sign of the sensor is detected using the calculated evaluation index.

JP 2011-230634 A

  However, the above-described conventional technology has room for further improvement in easily and accurately grasping a deviation from a normal state indicating a failure sign of a mechanical facility.

  Specifically, for example, when the machine facility is a large mechatronics machine (hereinafter referred to as “large machine”) such as a large refrigerator or a plant, the number of sensors becomes enormous. For this reason, when the above-mentioned conventional technology is applied to such a large machine, it is not difficult to imagine that the system will be complicated even if only the points that require the correspondence of the threshold value to each sensor are taken.

  In addition, although the above-described prior art is suitable for detecting a clear abnormal state exceeding a threshold value, a sensor that appears in the behavior of the entire system due to, for example, aging of each component, but exceeds the threshold value. It is unsuitable for detecting unclear abnormal states whose values are not shown. In other words, it is impossible to accurately capture a deviation from a normal state that indicates a sign of failure of the mechanical equipment.

  One aspect of the embodiment has been made in view of the above, and a failure sign determination method, a failure sign determination apparatus, and a failure sign determination apparatus capable of easily and accurately grasping a deviation from a normal state indicating a failure sign of mechanical equipment, and An object is to provide a failure sign determination program.

  The failure sign determination method according to an aspect of the embodiment includes a collection process, an extraction process, a setting process, a generation process, a calculation process, and a determination process. The collection step collects a data set including detection values of a plurality of sensors provided in the machine facility. The extracting step extracts a predetermined normal period during which the mechanical equipment is in a normal state and a past predetermined latest period from the data set. The setting step associates first additional information, which is additional information indicating that there is no failure, with the normal period, and unconditionally sets second additional information indicating that there is a failure with respect to the latest period. After the association, two groups in which the normal period and the latest period are mixed are set. The generation step generates a prediction model for predicting the additional information corresponding to input data by learning using one of the two groups. The calculation step calculates a probability that the additional information is predicted to be the second additional information in a prediction result of the prediction model obtained by using the other of the two groups as the input data. In the determination step, a failure sign of the mechanical equipment is determined based on the probability.

  According to one aspect of the embodiment, it is possible to easily and accurately capture a deviation from a normal state indicating a failure sign of a mechanical facility.

FIG. 1A is a schematic explanatory diagram (part 1) of the failure sign determination method according to the embodiment. FIG. 1B is a schematic explanatory diagram (part 2) of the failure sign determination method according to the embodiment. FIG. 1C is a schematic explanatory diagram (part 3) of the failure sign determination method according to the embodiment. FIG. 1D is a schematic explanatory diagram (part 4) of the failure sign determination method according to the embodiment. FIG. 1E is a schematic explanatory diagram (part 5) of the failure sign determination method according to the embodiment. FIG. 1F is a schematic explanatory diagram (No. 6) of the failure sign determination method according to the embodiment. FIG. 2A is a block diagram of the failure sign determination system according to the embodiment. FIG. 2B is a block diagram of the evaluation unit. FIG. 3A is an explanatory diagram (part 1) of the data shaping process by the extraction unit. FIG. 3B is an explanatory diagram (part 2) of the data shaping process performed by the extraction unit. FIG. 3C is an explanatory diagram of tag setting processing by the tag setting unit. FIG. 4A is a diagram (part 1) illustrating an example of an effect of the failure sign determination apparatus according to the embodiment. FIG. 4B is a diagram (part 2) illustrating an example of an effect of the failure sign determination apparatus according to the embodiment. FIG. 5 is a flowchart showing a processing procedure executed by the failure sign determination apparatus. FIG. 6 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the failure sign determination apparatus.

  Hereinafter, embodiments of a failure sign determination method, a failure sign determination device, and a failure sign determination 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.

  First, an outline of the failure sign determination method according to the present embodiment will be described with reference to FIGS. 1A to 1F. 1A to 1F are schematic explanatory views (No. 1) to (No. 6) of the failure sign determination method according to the embodiment.

  As illustrated in FIG. 1A, the target machine 100 includes a sensor group of sensors S-1 to Sn. In the present embodiment, the detection value from the sensor group is handled as one data set related to the operation of the target machine 100.

  In this way, by handling the detection values from the sensor group as a unit and evaluating them in units of data sets, the behavior of the entire target machine 100 including the correlation between the sensors S-1 to Sn at the time of operation can be obtained. It becomes possible to grasp the change. In addition, although the sensor group is mentioned here as an example, any component that can output an output value may be used.

  In the following figures, such data sets are indicated by circles as shown in FIG. 1A. Further, there are cases where the inside of such a circle is filled or a character is written in the circle, and this point will be described each time.

  As shown in FIG. 1B, in the present embodiment, such a data set is collected at a predetermined cycle during operation of the target machine 100, and among the collected data sets, “normal period” and “most recent period”. And used.

  Here, “for the normal period” refers to a predetermined period during which the target machine 100 is in a normal state in the initial operation stage. A large mechatronics machine such as the target machine 100 is normally expected to operate stably for several years from the initial operation, but during normal periods, environmental factors and individual difference factors of the target machine 100 are averaged. It is preferably set to about several tens of days from the initial operation. In this embodiment, it is assumed that the normal period is 30 days from the first operation.

  Further, “for the most recent period” refers to the most recent period in the past from the sign determination date for determining the sign of failure. In the present embodiment, it is assumed that the latest period is two weeks.

  In addition, among the data sets for “the most recent period” illustrated in FIG. 1B, the solid circles indicate data sets including the behavior of the target machine 100 different from the “normal period”. Such a data set will be described later with reference to FIG. 1F.

  In the present embodiment, the “normal period” and the “most recent period” are combined, and then divided into two groups in which “normal period” and “most recent period” are mixed, and one of them is divided into two groups. The other is used for “for generating a tag prediction model” and the other for “for evaluation using a tag prediction model”.

  Here, the “tag” will be described. As shown in FIG. 1C, in the present embodiment, an “N” (normal) tag indicating that there is no failure is assigned to each “normal period” data set. On the other hand, a tag “F” (failure) that is unconditionally assumed to be a failure is assigned to each data set of “the most recent period”.

  The tag of “F” given unconditionally to each data set of “the latest period” is alive in the generation of a tag prediction model 12d (see FIG. 1D) described later and the evaluation by the tag prediction model 12d. Come. The description will be continued sequentially. In the following, for tagged data sets, the letters “N” and “F” of the tags are described in circles and shown in the figure.

  As shown in FIG. 1D, in the present embodiment, tags corresponding to “normal period” or “most recent period” input by, for example, machine learning using each data set of the “generation” group described above. A tag prediction model 12d for predicting

  In this embodiment, a random forest is used as a machine learning algorithm. Since the random forest is publicly known, detailed description is omitted.

In this embodiment, the probability P _i that the tag is predicted to be “F” is calculated based on the prediction result of the tag prediction model 12 d to which each data set of the “for evaluation” group is input. .

Here, the probability P _i, as shown in FIG. 1E, for example, if "immediate period" and the "normal period" is short from sign determination date, originally differences due to the influence of such aging, "normal period" Are unlikely to appear between the current data set and the “most recent” data set.

In such a case, in the generated tag prediction model 12d, it is difficult to make a difference between the reference space of the “N” tag data set and the reference space of the “F” tag data set. Therefore, depending on the tag prediction model 12d in this case, the “N” tag data set may be predicted as the “F” tag, or the “F” tag data set may be predicted as the “N” tag. , out with a "normal period" and "the most recent period", the tag is hard out a difference in the probability P _i that is predicted to be "F".

  On the other hand, as shown in FIG. 1F, for example, when the “latest period” and the “normal period” are far from the sign determination date, “F” and the inside of the circle are filled with “behave different from the normal period”. As shown by the existence of the “including data set”, a difference due to an influence such as secular change tends to appear between the “normal period” data set and the “most recent period” data set.

In such a case, in the generated tag prediction model 12d, a difference easily occurs between the reference space of the “N” tag data set and the reference space of the “F” tag data set. Therefore, depending on the tag prediction model 12d in such a case, the possibility that the data set of the tag “F” is predicted to be “F” is high, so that the tag is “normal period” and “most recent period”. A difference appears in the probability P _i predicted to be “F”.

Returning to the description of FIG. 1D. In the present embodiment, it calculates the difference between the probability P _i of followed by "normal period" and "the most recent period" as deviance from a normal state. The degree of divergence is an indicator of how much the data set for “the most recent period” deviates from the data set for “normal period”. Therefore, by determining the degree of divergence based on, for example, a determination threshold value, it is possible to accurately grasp the deviation from the normal state indicating the failure sign.

  In the present embodiment, it is not necessary to associate a threshold value with the sensor value for each of the sensors S-1 to Sn. For this reason, it is possible to easily grasp a deviation from a normal state indicating a failure sign without complicating the system.

  In the present embodiment, the random forest is used to generate the tag prediction model 12d as described above. In this case, the contributions of the sensors S-1 to Sn are acquired from the tag prediction model 12d. be able to. Therefore, when it is determined that there is a failure sign based on the degree of divergence, it is the true cause of the failure sign depending on which of the sensors S-1 to Sn contributes to the failure sign. A specific site can be estimated. Thereby, the maintainability with respect to the target machine 100 can be improved.

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

  FIG. 2A is a block diagram of the failure sign determination system 1 according to the present embodiment. In FIG. 2A, constituent elements necessary for explaining the characteristics 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. 2A 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 an arbitrary unit according to various loads or usage conditions.・ It can be integrated and configured.

  In the description using FIG. 2A, the description of the components already described so far may be simplified or omitted. FIG. 2B is a block diagram of the evaluation unit 11e.

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

  The failure sign determination 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 tag setting unit 11c, a model generation unit 11d, an evaluation unit 11e, a determination unit 11f, and a notification unit 11g.

  The storage unit 12 is a storage device such as a hard disk drive, a nonvolatile memory, or a register, and includes collected data 12a, a normal period data group 12b, a latest period data group 12c, a tag prediction model 12d, and evaluation information 12e. Remember. The evaluation information 12e includes a probability 12ea, a divergence degree 12eb, and a contribution rate 12ec.

  The control unit 11 performs overall control of the failure sign determination apparatus 10. The collection unit 11a collects a data set from the sensor group of the target machine 100 at a predetermined period and stores it 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 secular change or the like.

  The extraction unit 11b extracts a data set for the normal period and a data set for the most recent period from the collected data 12a based on the normal period set at the first operation and the past most recent period from the sign determination date. They are stored in the normal period data group 12b and the latest period data group 12c, respectively.

  In addition, when extracting each data set, the extraction unit 11b can perform data shaping such as deleting unnecessary portions on the data in order to accurately determine the behavior of the target machine 100. This point will be specifically described with reference to FIGS. 3A and 3B. 3A and 3B are explanatory diagrams (No. 1) and (No. 2) of the data shaping process performed by the extraction unit 11b.

  As shown in the upper part of FIG. 3A, the extraction unit 11b performs a data set that is stopped for each of the normal period and the most recent period, that is, a data set when the target machine 100 is not operating or in an idling state. If there is, a thinning process is performed as shown in the lower part of FIG. 3A.

  Thus, in order to determine the failure sign, it is possible to cut a data set that can be a noise component in grasping the behavior of the target machine 100 during operation. For example, the prediction accuracy of the above-described tag prediction model 12d This can contribute to improving the accuracy of failure sign determination.

  Similarly, as shown in FIG. 3B, a portion that can be a noise component in one data set can be thinned out or the interval can be reduced. FIG. 3B is an example of detection data of the “A liquid flow velocity” detection sensor in the target machine 100, for example, but the level values other than the M1 portion shown in the upper part of FIG. 3B are portions corresponding to, for example, idling. The extraction unit 11b can remove portions other than the M1 portion.

  In addition, as shown in the lower part of FIG. 3B, the extraction unit 11b can pack a portion that is spaced apart in the horizontal axis direction in the M1 part. Also by this, data can be narrowed down to a portion showing the behavior of the target machine 100 during operation, which can contribute to improving the accuracy of failure sign determination.

  Returning to the description of FIG. 2A, the tag setting unit 11c will be described. The tag setting unit 11c associates (provides) a tag of “N” indicating that there is no failure with each data set stored in the normal period data group 12b. Further, the tag setting unit 11c unconditionally associates a tag “F” indicating failure with each data set stored in the latest period data group 12c.

  In addition, the tag setting unit 11c forms two groups in which the data sets for the normal period and the data sets for the latest period are mixed. These points will be specifically described with reference to FIG. 3C. FIG. 3C is an explanatory diagram of tag setting processing by the tag setting unit 11c.

  As illustrated in FIG. 3C, the tag setting unit 11c associates a tag “N” with each data set for the normal period. Further, the tag setting unit 11c unconditionally associates the tag “F” with each data set for the most recent period regardless of whether or not they include behavior different from that of the normal period.

  In addition, the tag setting unit 11c includes two groups of “N” tag data sets and “F” tag data sets, “generation” of the tag prediction model 12d, and tag prediction model 12d. Set “for evaluation” by.

  It should be noted that which data set is assigned to “for generation” or “for evaluation” in the normal period or the latest period may be random. Further, it is not necessary to distribute the data sets of the “N” and “F” tags so as to be equal in number to “for generation” and “for evaluation”.

  However, the number of data sets in the “generation” group may be equal to or greater than the number of data sets in the “evaluation” group so that the tag prediction model 12d has a certain degree of prediction accuracy. preferable. The ratio of the number of “N” and “F” tag data sets in one group is preferably equivalent to “number of“ N ”tag data sets> number of“ F ”tag data sets × 2” ”.

  Returning to the description of FIG. 2A, the model generation unit 11d will be described. The model generation unit 11d generates a tag prediction model 12d using each data set of the “generation” group formed in the tag setting unit 11c.

  The evaluation unit 11e inputs each data set of the “for evaluation” group formed in the tag setting unit 11c to the tag prediction model 12d and receives a prediction result by the tag prediction model 12d.

  And the evaluation part 11e calculates the various evaluation values for evaluating a failure sign based on the prediction result, and stores it in the evaluation information 12e. The evaluation value corresponds to the probability 12ea, the divergence degree 12eb, and the contribution rate 12ec.

  Here, the evaluation unit 11e will be described more specifically. As illustrated in FIG. 2B, the evaluation unit 11e includes a model input unit 11ea, a probability calculation unit 11eb, a divergence degree calculation unit 11ec, and a contribution rate calculation unit 11ed.

The model input unit 11ea inputs each data of the above-mentioned “for evaluation” group, which is shown as “a tagged data group” in the drawing, to the tag prediction model 12d. The probability calculation unit 11eb calculates the probability P _i that the tag is predicted to be “F” in each data set based on the prediction result output by the tag prediction model 12d. Further, the probability calculation section 11eb stores the calculated probability _{P i} to the probability 12ea of evaluation information 12e.

Deviation calculating unit 11ec, based on the probability 12Ea, calculates the difference between the probability P _i of the normal period and the last period as the divergence degree from the normal state.

Specifically, the deviation calculating unit 11ec takes out from the probability 12ea evaluation information 12e, for example, the minimum value of the probability P _i in the normal period, and a minimum value of the probability P _i the most recent period, these differences Is calculated as the degree of deviation from the normal state. Further, the divergence degree calculation unit 11ec stores the calculated divergence degree in the divergence degree 12eb of the evaluation information 12e.

The contribution rate calculation unit 11ed acquires the contributions of the sensors S-1 to Sn from the tag prediction model 12d, and calculates the sensors S-1 to S-1 from the expression “contribution rate _i = contribution _i / Σcontribution _i ”. The contribution rate for each Sn is calculated. In addition, the contribution rate calculation unit 11ed stores the calculated contribution rate in the contribution rate 12ec of the evaluation information 12e.

  Returning to the description of FIG. 2A, the determination unit 11f will be described. The determination unit 11f refers to the divergence degree 12eb, determines that there is a failure sign when the divergence degree from the normal state is equal to or greater than a predetermined determination threshold value, and issues a notification request instruction to the notification unit 11g.

  When the notification unit 11g receives a notification request instruction with a sign of failure from the determination unit 11f, the notification unit 11g notifies the external device of an alert notification. At this time, the notifying unit 11g refers to the divergence degree 12eb and the contribution rate 12ec, and indicates the divergence degree from the normal state, for example, the names of the sensors S-1 to Sn up to the fifth highest contribution rate. It can also be notified.

  4A and 4B show an example of the effect of the failure sign determination apparatus 10 according to the present embodiment. 4A and 4B are diagrams (part 1) and (part 2) illustrating an example of an effect of the failure sign determination apparatus 10 according to the embodiment.

  4A and 4B, in order to make the explanation easy to understand, when the failure sign determination apparatus 10 is caused to perform a simulation operation based on the operation data of the target machine 100 in which a failure actually occurred on the “failure occurrence date”. The figure is shown.

  Specifically, in FIG. 4A, a failure has occurred in the target machine 100 on the “failure occurrence date” on the right end of the horizontal axis, but the failure sign determination device 10 performs a failure sign determination process every day in the past of the “failure occurrence date”. It is a simulation result at the time of assuming that it was performed. The determination threshold value for determining the degree of divergence was set to about 0.15.

Then, first, the deviation from the normal state calculated by the failure sign determination device 10 is not plotted on the graph from the cumulative operation days “−160” to “−120”. This has little difference in the probability P _i in a normal period and the last period, i.e., "- 160" in the "-120" day to day, the behavior of the entire objective machine 100, the deviation from the normal state almost It means not.

  However, the degree of divergence starts to increase little by little from the cumulative operating days “−120” days, and the degree of divergence is greater than or equal to the determination threshold value in the vicinity of “−90” days (see “Predictable detection days” in the figure). From this, it can be seen that a failure sign can be detected 90 days before the “failure occurrence date”. It can also be seen that the degree of divergence has increased rapidly since the “predictable detection date”. Therefore, if the failure sign determination process by the failure sign determination device 10 according to the present embodiment is performed, it is understood that the failure on the current “failure occurrence date” can be prevented.

  FIG. 4B shows the top five rankings of the contribution ratios of the sensors S-1 to Sn in the “predictable detection date” and the “failure occurrence date” in FIG. 4A. From this, it can be seen that the sensors S-1 to Sn having a high contribution rate are different in each of the “predictable detection date” and the “failure occurrence date”.

  This shows that factors that may lead to failure can also be changed day by day due to a gradual change in the correlation between the interlocked sensors S-1 to S-n due to aging. In addition, since no countermeasure was taken for each of the sensors S-1 to Sn having a higher contribution rate to the “predictable detection date”, another sensor S-1 was selected for the “failure occurrence date”. It is also possible that ~ Sn is affected.

  Therefore, if failure sign determination processing by the failure sign determination device 10 is performed in the present embodiment, measures should be taken for each of the sensors S-1 to Sn having a high contribution rate at the time of “predictable detection date”. Therefore, it can be seen that the failure on this “failure occurrence date” could be prevented.

  Next, a processing procedure executed by the failure sign determination apparatus 10 will be described with reference to FIG. FIG. 5 is a flowchart illustrating a processing procedure executed by the failure sign determination apparatus 10.

  As shown in FIG. 5, 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 performs an initial information setting process (step S102).

  In the initial information setting process, for example, the number of days in the normal period or the latest period (in this embodiment, “30 days” and “2 weeks” in order) is set on the system. If it is not the first operation (No in step S101), the control is transferred to step S103.

  Subsequently, the extraction unit 11b extracts each data set for the normal period and each data set for the most recent period from the sign determination date from the collected data 12a collected by the collection unit 11a (step S103). At this time, the extraction unit 11b performs data shaping processing for each extracted data set so that only the portion necessary for failure sign determination is included.

  Subsequently, for example, the tag setting unit 11c determines whether or not the number of data sets for the most recent period is equal to or greater than the necessary minimum number (step S104). For example, in the present embodiment, the necessary minimum number here is about 140.

  Here, when the determination condition of step S104 is satisfied (step S104, Yes), it is subsequently determined whether or not the number of data sets on the day of the sign determination date is equal to or greater than the necessary minimum number (step S105). For example, in this embodiment, the necessary minimum number here is about 10.

  In addition, when the determination condition of step S104 or S105 is not satisfied (step S104, No / step S105, No), the fact that each necessary minimum number is not satisfied affects the prediction accuracy of the generated tag prediction model 12d. Therefore, the process is terminated.

  On the other hand, when the determination condition of step S105 is satisfied (step S105, Yes), the tag setting unit 11c subsequently sets the tag “N” for each data set for the normal period and each data set for the latest period. The tag “F” is assigned to each (step S106).

  Subsequently, the tag setting unit 11c groups the tagged data into two groups so that the “N” tag data set and the “F” tag data set are mixed, respectively (step S107).

  Then, the model generation unit 11d generates the tag prediction model 12d using one of the “generation” data sets that has been grouped (step S108).

Subsequently, the evaluation unit 11e inputs the other grouped “for evaluation” data set to the tag prediction model 12d, and the tag is predicted to be “F” based on the prediction result returned by the tag prediction model 12d. Probability P _i is calculated (step S109).

The evaluation unit 11e is taken out and a minimum value of the probability P _i in the normal period, and a minimum value of the probability P _i the most recent period, and calculates these differences as deviance from a normal state (Step S110) . Here, not necessarily the minimum value of the probability P _i, may be an average value. That is, it may be a representative value.

  And the evaluation part 11e calculates the contribution rate of each sensor S1-Sn (step S111). Subsequently, the determination unit 11f determines whether or not the degree of divergence calculated by the evaluation unit 11e is greater than or equal to a predetermined determination threshold value (step S112).

  Here, when the degree of divergence is greater than or equal to the determination threshold value (step S112, Yes), the notification unit 11g notifies a failure sign (step S113) and ends the process. Further, when the determination condition of step S112 is not satisfied (step S112, No), the process ends. In this case, the notifying unit 11g may notify that there is no sign of failure.

  Note that the failure sign determination apparatus 10 according to the embodiment 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 function of the failure sign determination 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 corresponds to a communication unit (not shown) with the target machine 100, receives data from other devices via the communication network, sends the data to the CPU 61, and transmits the data generated by the CPU 61 via the communication network. Send to other devices.

  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.

  When the computer 60 functions as the failure sign determination device 10, the CPU 61 of the computer 60 executes a program loaded on the RAM 62, thereby collecting the collection unit 11a, the extraction unit 11b, the tag setting unit 11c, the model generation unit 11d, The functions of the evaluation unit 11e, the determination unit 11f, and the notification unit 11g are realized. The HDD 64 realizes the function of the storage unit 12 and stores collected data 12a and the like.

  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 sign determination apparatus 10 according to the embodiment includes the collection unit 11a, the extraction unit 11b, the tag setting unit 11c (corresponding to an example of “setting unit”), and the model generation unit 11d (“generation”). An evaluation unit 11e (corresponding to an example of “calculation unit”), and a determination unit 11f.

  The collection unit 11a collects a data set including detection values of a plurality of sensors S-1 to Sn provided in the target machine 100 (corresponding to an example of “mechanical equipment”). The extraction unit 11b extracts a predetermined normal period in which the target machine 100 is in a normal state and a past predetermined latest period from the data set.

  The tag setting unit 11c associates the “N” tag (corresponding to an example of “first additional information”), which is a tag indicating that there is no failure (corresponding to an example of “additional information”), with the normal period. In addition, the “F” tag (corresponding to an example of “second additional information”) indicating a failure is unconditionally associated with the latest period, and the normal period and the latest period are mixed. 2 groups are set.

  The model generation unit 11d generates a tag prediction model 12d (corresponding to an example of a “prediction model”) that predicts a tag corresponding to input data by learning using one of the two groups.

The evaluation unit 11e calculates a probability P _i that the tag is predicted to be a tag of “F” in the prediction result of the tag prediction model 12d obtained by using the other of the two groups as input data. The determination unit 11f determines a failure sign of the target machine 100 based on the probability P _i .

  Therefore, according to the failure sign determination device 10 according to the present embodiment, it is possible to easily and accurately grasp the deviation from the normal state indicating the failure sign of the target machine 100.

  In the above-described embodiment, a random forest is used as the machine learning algorithm, but the machine learning method is not limited. Accordingly, the tag prediction model 12d may be generated by executing machine learning by a regression analysis method such as support vector regression using a pattern classifier such as SVM (Support Vector Machine). Here, the pattern discriminator is not limited to SVM, and may be, for example, AdaBoost. Further, deep learning 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 sign determination system 10 Failure sign determination apparatus 11 Control part 11a Collection part 11b Extraction part 11c Tag setting part 11d Model generation part 11e Evaluation part 11ea Model input part 11eb Probability calculation part 11ec Deviation degree calculation part 11ed Contribution rate calculation part 11f Determination Unit 11g Notification unit 12 Storage unit 12a Collected data 12b Normal period data group 12c Latest period data group 12d Tag prediction model 12e Evaluation information 100 Target machine S-1 to Sn sensor

Claims (9)

A collection step of collecting a data set including detection values of a plurality of sensors provided in the mechanical facility;
An extraction step for extracting a predetermined normal period in which the mechanical equipment is in a normal state and a predetermined past period in the past from the data set;
The first additional information, which is additional information indicating that there is no failure, is associated with the normal period, and the second additional information that indicates failure is unconditionally associated with the latest period. A setting step for setting two groups in which the normal period and the latest period are mixed;
A generation step of generating a prediction model for predicting the additional information corresponding to input data by learning using one of the two groups;
A calculation step of calculating a probability that the additional information is predicted to be the second additional information in a prediction result of the prediction model obtained by using the other of the two groups as the input data;
And a determination step of determining a failure sign of the mechanical equipment based on the probability. The calculation step includes
The difference between the representative values of the probabilities in each of the normal period and the latest period is calculated as the degree of deviation from the normal state of the mechanical equipment,
The determination step includes
The failure sign determination method according to claim 1, wherein, when the degree of deviation is equal to or greater than a predetermined determination threshold value, it is determined that the machine facility has a failure sign. The failure sign determination method according to claim 2, wherein the representative value is a minimum value of the probability in each of the normal period and the latest period. The calculation step includes
4. The failure sign according to claim 1, wherein a contribution degree of each of the sensors to the prediction result is obtained from the prediction model, and a contribution rate of each of the sensors is calculated based on the contribution degree. Judgment method. A notification step of performing an alert notification to an external device when it is determined by the determination step that there is a failure sign,
The determination step includes
5. The failure sign according to claim 4, wherein, when it is determined that the failure sign is present, information on the sensor having a high contribution rate calculated by the calculation step is included in the alert notification of the notification step. Judgment method. The normal period for the normal period is:
The period from the initial operation of the mechanical equipment to the time when environmental factors and individual difference factors of the mechanical equipment are assumed to be averaged is set. The failure sign determination method described in 1. The extraction step includes
The failure sign according to any one of claims 1 to 6, wherein data shaping that removes a data portion corresponding to an idling state of the mechanical equipment is performed for each of the normal period and the most recent period. Judgment method. A collection unit for collecting a data set including detection values of a plurality of sensors provided in the mechanical facility;
Among the data set, an extraction unit that extracts a predetermined normal period in which the mechanical equipment is in a normal state and a past predetermined latest period;
The first additional information, which is additional information indicating that there is no failure, is associated with the normal period, and the second additional information that indicates failure is unconditionally associated with the latest period. A setting unit for setting two groups in which the normal period and the latest period are mixed;
A generation unit that generates a prediction model for predicting the additional information corresponding to input data by learning using one of the two groups;
A calculation unit that calculates a probability that the additional information is predicted to be the second additional information in the prediction result of the prediction model obtained by using the other of the two groups as the input data;
A failure sign determination apparatus comprising: a determination unit that determines a failure sign of the mechanical equipment based on the probability. On the computer,
A collection procedure for collecting a data set including detection values of a plurality of sensors provided in a mechanical facility;
An extraction procedure for extracting, from the data set, a predetermined normal period in which the mechanical equipment is in a normal state and a predetermined past period in the past;
The first additional information, which is additional information indicating that there is no failure, is associated with the normal period, and the second additional information that indicates failure is unconditionally associated with the latest period. A setting procedure for setting two groups in which the normal period and the latest period are mixed,
A generation procedure for generating a prediction model for predicting the additional information corresponding to input data by learning using one of the two groups;
A calculation procedure for calculating a probability that the additional information is predicted to be the second additional information in the prediction result of the prediction model obtained by using the other of the two groups as the input data;
And a determination procedure for determining a failure sign of the mechanical equipment based on the probability.

Images