WO2016067852A1 - Diagnostic job generating system, diagnostic job generating method, and diagnostic job generating display method - Google Patents

Abstract

To carry out efficient diagnostic job generation, provided is a diagnostic job generating system, comprising: a diagnostic model generating unit (112) which generates a plurality of diagnostic models (Md) corresponding to types by generating combinations of diagnostic processing units (Sw) formed from a plurality of processes and generating a plurality of common conditions (Pc) which can be applied to devices (2) of the same type; a diagnostic trial run unit (113) and a diagnostic performance determination unit (114) which extract, from the plurality of generated diagnostic models (Md), the diagnostic model (Md) with the highest diagnostic performance; and a diagnostic job generating unit (115) which, by applying individual conditions (Pm) of each of the devices (2) to the extracted diagnostic model (Md), generates a diagnostic job (122) for each of the devices (2).

Description

Diagnostic job generation system, diagnostic job generation method, and diagnostic job generation display method

The present invention relates to a technique for a diagnostic job generation system, a diagnostic job generation method, and a diagnostic job generation display method related to a predictive diagnosis of equipment.

Conventionally, the main maintenance of equipment is “Time-Based Maintenance (TBM)” based on the operation time of the equipment. However, with the development of sensing technology for device operation status, “condition-based maintenance (CBM)” based on the device operation status is becoming widespread. The CBM monitors the state of the device using a diagnostic statistical model (hereinafter referred to as a diagnostic model) according to the operating state of the device, and prompts early maintenance of the system by issuing a warning or the like when it is determined to be abnormal. At this time, it is necessary to pay attention to how individual differences of devices are incorporated into the diagnosis model.

As such a technique, the technique described in Patent Document 1 is disclosed. The technique described in Patent Document 1 generates individual regression equations for the remaining N−1 items when a certain device A is a diagnosis target among a plurality (N) of similar devices. And the technique of patent document 1 produces | generates a meta model from the group of the produced | generated regression equation, and applies the sensor data of the apparatus A of a diagnostic object with respect to the produced | generated meta model. By doing in this way, the technique of patent document 1 obtains the diagnostic regression formula of the apparatus A to be diagnosed.

As described above, the technique described in Patent Document 1 uses a device similarity to quickly add a small number of newly installed devices to a diagnosis target based on information obtained from a large number of existing devices.

Japanese Patent No. 5297272

However, there are cases where the mass production scale of the device to be diagnosed is small or sufficient information cannot be obtained from the existing device. Even if the mass production scale is large, sufficient information cannot be obtained from existing equipment because a sufficient number is not on the market at the beginning of the sale. Therefore, in the technique described in Patent Document 1, when a diagnostic model corresponding to a large number of new devices is generated from a small number of existing devices, there is little information obtained from the small number of existing devices. It is necessary to estimate the information for each of a large number of newly installed devices. Therefore, the man-hour for producing | generating the diagnostic model corresponding to many new installations will increase.

The present invention has been made in view of such a background, and an object of the present invention is to efficiently generate a diagnostic job.

In order to solve the above-described problem, the present invention provides a diagnosis model generation unit that generates a plurality of common diagnosis models applicable to devices belonging to the same group, and a diagnosis that evaluates the performance of the plurality of generated diagnosis models. An extraction unit for extracting a predetermined diagnosis model from the plurality of generated diagnosis models based on the diagnosis result, and each of the devices for the extracted diagnosis model. And a diagnostic job generation unit that generates a diagnostic job for each device by applying an individual condition that is a condition corresponding to the above.

According to the present invention, an efficient diagnostic job can be generated.

It is a figure showing an outline of composition of a diagnostic job generation system concerning a 1st embodiment. It is a figure which shows the outline of an internal structure of the diagnostic job generation apparatus which concerns on 1st Embodiment. It is a figure which shows the hardware structural example of each apparatus which comprises the diagnostic job generation system which concerns on 1st Embodiment. It is a figure which shows the outline | summary of the change process of the diagnostic process part which concerns on 1st Embodiment. It is a figure which shows the outline | summary of the change process of the common condition which concerns on 1st Embodiment. It is a figure which shows the outline | summary of the change process of the individual condition which concerns on 1st Embodiment. It is a schematic diagram which shows the combination of the diagnostic process part in a diagnostic model, a common condition, and an individual condition. It is a flowchart which shows the procedure of the diagnostic trial process which concerns on 1st Embodiment. It is a flowchart which shows the procedure of the diagnostic performance determination process which concerns on 1st Embodiment. It is a figure which shows the result of the diagnostic performance determination process which concerns on 1st Embodiment. It is a figure which shows the example of a diagnostic job template and a diagnostic job, (a) is a figure which shows the example of the diagnostic job template which concerns on 1st Embodiment, (b) is the diagnosis which concerns on 1st Embodiment. It is a figure which shows the example of a job. It is a flowchart which shows the procedure of the diagnostic job production | generation process which concerns on 1st Embodiment. It is a figure which shows the example of the diagnostic job which concerns on 1st Embodiment, (a) shows the example of the diagnostic job of No. 2 machine, (b) shows the example of the diagnostic job of No. 3 machine, (c) shows the No. N machine Of an example diagnostic job It is a figure which shows the example of the diagnostic job list screen which concerns on 1st Embodiment. It is a figure which shows the diagnostic model which concerns on 2nd Embodiment.

Next, modes for carrying out the present invention (referred to as “embodiments”) will be described in detail with reference to the drawings as appropriate.

[First Embodiment]
(System overview)
FIG. 1 is a diagram showing an outline of the configuration of the diagnostic job generation system according to the first embodiment. In FIG. 1, broken line arrows indicate data flow, and alternate long and short dash lines indicate data correspondence.
The diagnostic job generation system 10 includes a plurality of devices 2, the diagnostic job generation device 1, and a display device 350 on which a diagnostic job list screen is displayed.
The devices 2 are to be diagnosed by the diagnostic job 122 generated by the diagnostic job generation system 10, and each device 2 has model information common to a plurality of devices 2 and an ID (Identification) unique to each device 2. ). In the present embodiment, each device 2 has the same model “A”.

The diagnostic job generation apparatus 1 includes a diagnostic model generation unit 112, a diagnostic model Md, a diagnostic job generation unit 115, and a plurality of diagnostic jobs 122. The diagnosis job 122 corresponds to each device 2. Since the internal configuration of the diagnostic job generation apparatus 1 will be described later, only the main part will be described here.

The diagnostic model generation unit 112 generates a diagnostic model Md.
The diagnostic model Md has a diagnostic processing unit Sw, a common condition Pc, and an individual condition Pm. The diagnosis processing unit Sw and the common condition Pc are associated with the model of the device 2 in 1: 1, and the individual condition Pm is associated with each device 2 in 1: 1. The common condition Pc is a condition common to the same type of devices 2, and the individual condition Pm is an independent diagnosis condition for each device 2.

The diagnosis job generation unit 115 generates a diagnosis job 122 from the diagnosis model Md.
The display device 350 displays a diagnostic job list screen 500 (FIG. 14) on which information about the generated diagnostic job 122 is displayed.

FIG. 2 is a diagram illustrating an outline of an internal configuration of the diagnostic job generation apparatus according to the first embodiment. In FIG. 2, solid arrows indicate the flow of processing, and broken arrows indicate the flow of data.
As illustrated in FIG. 2, the diagnostic job generation apparatus 1 includes history information 121, a diagnostic trial unit (extraction unit) 113, a diagnostic performance determination unit (extraction unit) 114, and a diagnostic job generation unit 115.
The diagnostic model Md is generated by the diagnostic model generation unit 112 as shown in FIG. 1, and includes a diagnostic processing unit Sw, a common condition Pc, and an individual condition Pm.

The diagnostic processing unit Sw has an input unit Si, a learning unit St, and a diagnostic unit Sd.
The common condition Pc has an input condition Pi that is an operating condition of the input unit Si, a learning condition Pt that is an operating condition of the learning unit St, and a diagnostic condition Pd that is an operating condition of the diagnostic unit Sd.
The individual condition Pm has a code book CB. The code book CB is generated by the learning unit St, and defines the operating condition of the diagnostic unit Sd together with the diagnostic condition Pd. That is, the code book CB stores the learning result by the learning unit St. The code book CB will be described later. The code book CB at the time of generating the diagnostic model Md is not the code book CB for all the devices 2 having the same model, but the device 2 of the devices 2 having the same model as a representative device. A code book CB for the device 2 is generated. By doing in this way, the diagnostic model Md becomes a common diagnostic model Md applicable to devices belonging to the same group.
Further, the diagnostic model generation unit 112 (see FIG. 1), which will be described later, the input unit Si of the diagnostic processing unit Sw, the learning unit St, and the diagnostic unit Sd are variously changed, and the input condition Pi of the common condition Pc. Alternatively, a plurality of diagnostic processing units Sw are generated by changing the learning condition Pt and the diagnostic condition Pd in various ways.

The history information 121 holds a plurality of histories H that are change histories of the diagnostic processing unit Sw, the common condition Pc, and the individual condition Pm that constitute the diagnostic model Md (FIG. 1).
The diagnosis trial unit 113 tries diagnosis for each combination of the history H managed by the history information 121 (that is, the diagnosis model Md).
The diagnostic performance determination unit 114 evaluates the diagnostic trial result tried by the diagnostic trial unit 113, searches for and extracts the combination of the history H with the best diagnostic performance (that is, the diagnostic model Md with the best diagnostic performance). .
The diagnosis job generation unit 115 generates a diagnosis job template from the diagnosis model Md selected by the diagnosis performance determination unit 114 and then generates a plurality of diagnosis jobs 122 corresponding to a large number of devices 2.

(Hardware configuration diagram)
FIG. 3 is a diagram illustrating an example of a hardware configuration of each device configuring the diagnostic job generation system according to the first embodiment. Reference is made to FIGS. 1 and 2 as appropriate.
The diagnostic job generation system 10 includes a diagnostic job generation device 1, a terminal device 3, and each device 2 to be diagnosed.
The diagnostic job generation apparatus 1 is an apparatus that generates a diagnostic job 122 corresponding to each device 2 as described above, and includes a memory 110, a storage device 120 such as an HD (Hard Disk), a CPU (Central Processing Unit) 130, and transmission / reception. A device 140 is included.
The program stored in the storage device 120 is expanded in the memory 110, and the program expanded in the memory 110 is executed by the CPU 130, whereby the processing unit 111, the diagnostic model generation unit 112 constituting the processing unit 111, the diagnosis A trial unit 113, a diagnostic performance determination unit 114, and a diagnostic job generation unit 115 are embodied.

The processing unit 111 controls the diagnostic model generation unit 112, the diagnostic trial unit 113, the diagnostic performance determination unit 114, and the diagnostic job generation unit 115.
The diagnosis model generation unit 112 generates a plurality of diagnosis processing units Sw, common conditions Pc, and individual conditions Pm that constitute the diagnosis model Md, and stores each of the generated diagnosis processing units Sw, common conditions Pc, and individual conditions Pm as a history H ( 2), the history information 121 of the storage device is stored.
The diagnosis trial unit 113, the diagnosis performance determination unit 114, and the diagnosis job generation unit 115 have already been described with reference to FIG.
The transmission / reception device 140 is connected to the device 2 and the terminal device 3 and transmits / receives information to / from these devices.

The terminal device 3 is a device that displays information related to the diagnostic model Md generated by the diagnostic job generation device 1 and the diagnostic job 122, and includes a memory 310, a storage device 320 such as an HD, a CPU 330, a display device 350 such as a display, and transmission / reception. A device 340 is included.
A program stored in the storage device 320 is expanded in the memory 310, and the program expanded in the memory 310 is executed by the CPU 330, thereby realizing the processing unit 311 and the display processing unit 312 constituting the processing unit 311. Has been.

The display processing unit 312 displays information related to the diagnostic model Md generated by the diagnostic job generation device 1 and the diagnostic job 122 on the display device 350. Note that the display device 350 of the terminal device 3 corresponds to the display device 350 of FIG.
The transmission / reception device 340 is connected to the diagnostic job generation device 1 and the device 2 and transmits / receives information to / from these devices.

The diagnostic job generation device 1, each device 2, and the terminal device 3 are connected to each other via a LAN (Local Area Network), a cloud network, or the like.

Also, each of the units 112 to 115 and the display processing unit 312 may be mounted on an integrated device, or a device on which at least one function is mounted may be installed.

(Diagnosis model generation)
Next, an overview of the diagnostic model Md generation processing by the diagnostic model generation unit will be described with reference to FIGS.
FIG. 4 is a diagram showing an outline of change processing of the diagnosis processing unit according to the first embodiment.
Each of the plurality of histories H held in the history information 121 is associated with a diagnostic processing unit Sw, and any of the processing steps of the input unit Si, the learning unit St, and the diagnostic unit Sd included in the diagnostic processing unit Sw. Is changed, the history H is updated (see FIG. 4). Similarly, when any of the input condition Pi, the learning condition Pt, and the diagnosis condition Pd in the common condition Pc is changed, the history H is updated (described later in FIG. 5). Further, when the code book CB of the individual condition Pm is changed, the history H is updated (described later in FIG. 6).
In the example shown in FIG. 4, the diagnostic model generation unit 112 (FIG. 3) first has a diagnostic model processing unit Sw1 (Sw) having an input unit Si1 (Si), a learning unit St1 (St), and a diagnostic unit Sd1 (Sd). Is generated. The diagnosis model generation unit 112 holds the diagnosis processing unit Sw1 in the history H1 (H).

Further, the diagnostic model generation unit 112 generates a diagnostic processing unit Sw2 (Sw) in which the input unit Si1 is changed to the input unit Si2 (Si) among the elements constituting the diagnostic model processing unit Sw1 stored in the history H1. To do. Then, the diagnosis model generation unit 112 holds the diagnosis processing unit Sw2 in the history H2 (H). Changing the input unit Si1 to the input unit Si2 means changing a sensor to be input.

Further, the diagnostic model generation unit 112 (see FIG. 3), among the elements constituting the diagnostic processing unit Sw2 stored in the history H2, the diagnostic processing unit Sw3 (the learning unit St2 is changed to the learning unit St2 (St)). Sw) is generated. Then, the diagnosis model generation unit 112 holds the generated diagnosis processing unit Sw3 in the history H3 (H). Changing the learning unit St is, for example, changing to the learning unit St2 that performs learning processing using a neural network if the learning unit St1 performs learning processing that performs clustering.

FIG. 5 is a diagram showing an outline of the common condition changing process according to the first embodiment.
A common condition Pc is associated with each history H in the history information 121, and any one of the operation conditions of the input condition Pi, the learning condition Pt, and the diagnosis condition Pd included in the common condition Pc is changed by the diagnosis model generation unit 112. When it is done, the history H is updated.
In the example illustrated in FIG. 5, the diagnosis model generation unit 112 first generates a common condition Pc1 (Pc) having an input condition Pi1 (Pi), a learning condition Pt1 (Pt), and a diagnosis condition Pd1 (Pd). Then, the diagnostic model generation unit 112 holds the generated Pc1 in the history H11 (H).

Next, the diagnosis model generation unit 112 generates a common condition Pc2 (Pc) in which the learning condition Pt1 is changed to the learning condition Pt2 (Pt) among the elements constituting the common condition Pc1 held in the history H11. Then, the diagnostic model generation unit 112 holds the generated common condition Pc2 in the history H12 (H). Changing the learning condition Pt means changing a learning parameter when performing the learning process.

Further, the diagnosis model generation unit 112 generates a common condition Pc3 (Pc) in which the diagnosis condition Pd1 is changed to the diagnosis condition Pd2 (Pd) among the elements constituting the common condition Pc2 held in the history H12. Then, the diagnostic model generation unit 112 holds the generated common condition Pc3 in the history H13 (H). The diagnosis condition Pd is, for example, diagnosis timing, a threshold value for reporting, and the like.

FIG. 6 is a diagram showing an overview of the individual condition changing process according to the first embodiment.
The history H in the history information 121 is associated with the individual condition Pm, and the history H is updated when the code book CB included in the individual condition Pm is changed by the diagnostic model generation unit 112.
In the example shown in FIG. 6, the diagnostic model generation unit 112 (see FIG. 3) generates the individual condition Pm1 (Pm) by generating the code book CB1 (CB). Then, the diagnostic model generation unit 112 holds the generated individual condition Pm1 in the history H21 (H).

Next, the diagnostic model generation unit 112 generates an individual condition Pm2 (Pm) in which the code book CB1 of the individual condition Pm1 held in the history H21 is changed to the code book CB2 (CB). Then, the diagnostic model generation unit 112 holds the generated individual condition Pm2 in the history H22 (H).

Furthermore, the diagnostic model generation unit 112 generates an individual condition Pm3 (Pm) in which the code book CB2 of the individual condition Pm2 held in the history H22 is changed to the code book CB3 (CB). Then, the diagnostic model generation unit 112 holds the generated individual condition Pm3 in the history H23 (H).
Since the contents of the code book CB are determined by the input data, the configuration of the learning unit St, and the learning condition Pt, the diagnostic model generation unit 112 is configured for each combination of the diagnostic processing unit Sw and the common condition Pc generated so far. Alternatively, the code book CB may be generated.

As described above, the code book CB at the time of generating the diagnostic model Md is not the code book CB for all the devices 2 having the same model, but a certain device 2 among the devices 2 having the same model. As a representative device, a code book CB for the device 2 is generated. Here, it is assumed that the code book CB of “No. 1” among the plurality of devices 2 is generated. Therefore, the input data input to the input unit Si is the input data of “No. 1”. The generation of the code book CB for the other device 2 will be described later.

In this embodiment, the combination of elements constituting the diagnostic processing unit Sw, the common condition Pc, and the individual condition Pm is held as the history H, but the diagnostic processing unit Sw, the common condition Pc, and the individual condition Pm are stored. The collected diagnosis model Md may be held in the history H.

(Combination of diagnostic models)
FIG. 7 is a schematic diagram illustrating a combination of a diagnosis processing unit, common conditions, and individual conditions in a diagnosis model. Note that broken line arrows in FIG. 7 indicate version transitions.
In FIG. 7, the combination of the diagnostic process Sw, the common condition Pc, and the individual condition Pm is shown in the form of a cube 400.
As shown in FIG. 4, the change of the diagnostic processing unit Sw is held as the history H. That is, in the example of FIG. 7, the diagnosis processing units Sw1 to Sw3 (Sw) are held. In FIG. 7, a surface 401 viewed from above the cube 400 corresponds.

Next, as shown in FIG. 5, the change of the common condition Pc is held as the history H. That is, in the example of FIG. 7, the common conditions Pc1 to Pc3 (Pc) and the common condition Pc4 (Pc) generated by changing a part of the common condition Pc1 (for example, the learning condition Pt) are retained. In FIG. 7, the upper half symbol 402 corresponds to the surface of the cube 400 viewed from the front side.

Further, as shown in FIG. 6, the change of the individual condition Pm is held as the history H. That is, in the example of FIG. 7, the individual conditions Pm1 to Pm3 (Pm) and the individual condition Pm4 (Pm) generated by changing the code book CB1 of the individual condition Pm1 are retained. In FIG. 7, the lower half reference numeral 403 corresponds to the surface of the cube 400 viewed from the front side.

Furthermore, as shown in FIGS. 4 to 5, the diagnosis processing unit Sw, the common condition Pc, and the individual condition Pm are held in a form linked to each other via the history H. For example, if the learning unit St performs a learning process using clustering, the learning condition Pt in the common condition Pc becomes the learning condition Pt for clustering and does not become the learning condition Pt for the neural network. In other words, the diagnosis processing unit Sw in which the learning unit St performs clustering and the common condition Pc having the learning condition Pt of the neural network cannot be combined (cannot be linked).

Further, if the learning unit St performs learning processing using clustering, the codebook CB of the individual condition Pm becomes a learning result by clustering, and does not become a learning result of the neural network. In other words, the individual condition Pm having the learning result of the neural network and the diagnosis processing unit Sw having the learning unit St for clustering and the common condition Pc having the learning condition Pt for clustering cannot be combined (cannot be linked). It is.
Thus, the configuration of the diagnostic processing unit Sw, the configuration of the common condition Pc, and the configuration of the individual condition Pm are mutually related.

Accordingly, among the combinations that can be combined in the cube 400, a combination of the diagnostic processing unit Sw that can be combined, the common condition Pc, and the individual condition Pm becomes the diagnosis model Md. For example, referring to the examples of FIGS. 4 to 6, {history H1, history H11, history H21} = {diagnostic processing unit Sw1, common condition Pc1, individual condition Pm1}, {history H2, history H12, history H22} = {Diagnostic processing unit Sw2, common condition Pc2, individual condition Pm2}, {history H3, history H13, history H23} = {diagnostic processing unit Sw3, common condition Pc3, individual condition Pm3} can be combined with diagnostic processing unit Sw It is assumed that the combination is a common condition Pc and an individual condition Pm.
In this way, the diagnostic model generation unit 112 generates a diagnostic model Md applicable to devices belonging to the same type (group).

In this embodiment, in FIGS. 4 to 6, the diagnosis processing unit Sw, the common condition Pc, and the individual condition Pm are individually changed. However, the present invention is not limited to this, and the diagnosis processing unit Sw, the common condition Pc, and The individual condition Pm may be changed in conjunction. For example, if the learning unit St in the diagnosis processing unit Sw is a learning method using a neural network, the format of the learning condition Pt in the common condition Pc is determined. Therefore, the diagnostic model generation unit 112 may change the learning condition Pt in a format that conforms to the learning method in the learning unit St of the diagnostic processing unit Sw. Similarly, if the learning process performed by the learning unit St and the learning condition Pt are determined, the code book CB of the individual condition Pm is also determined, so that the diagnostic model generation unit 112 has the input unit Si, the learning unit St, and the learning condition Pt. Based on the above, the code book CB may be generated.

If the learning process in the learning unit St is determined, the format of the input data is also determined. Therefore, the diagnostic model generation unit 112 may change the input unit Si and the input condition Pi in accordance with the learning process in the learning unit St. .

(Diagnosis trial process)
FIG. 8 is a flowchart showing the procedure of the diagnostic trial process according to the first embodiment. Reference is made to FIGS. 2 and 3 as appropriate.
First, the diagnosis trial unit 113 selects one combination of the history H (that is, a possible combination of the diagnosis processing unit Sw, the common condition Pc, and the individual condition Pm) (S101).
Then, the diagnostic trial unit 113 tries to diagnose a diagnostic model Md that is a combination of the selected diagnostic processing unit Sw, common condition Pc, and individual condition Pm (step S102). Specifically, the diagnostic trial unit 113 of the diagnostic model Md uses the diagnostic model Md for the device 2 to actually perform the above-mentioned diagnostic trial (diagnosis) by the CBM, and the result of the diagnosis is diagnostic result information. As a primary save.

Next, the diagnostic trial unit 113 determines whether or not the diagnostic trial has been completed for all combinations of the diagnostic processing unit Sw, the common condition Pc, and the individual condition Pm (S103).
As a result of step S103, when the trial of diagnosis is not completed for all combinations (S103 → No), the diagnostic trial unit 113 returns the process to step S101.
As a result of step S103, when the trial of diagnosis has been completed for all combinations (S103 → Yes), the diagnostic trial unit 113 passes information on the diagnostic trial result in each combination to the diagnostic performance determination unit 114 (step S104). ).

(Diagnosis performance determination process)
FIG. 9 is a flowchart showing the procedure of the diagnostic performance determination process according to the first embodiment.
First, the diagnostic performance determination unit 114 selects one diagnosis result (diagnosis trial result) tried by the diagnosis trial unit 113 (step S201).
Then, the diagnostic performance determination unit 114 plots a ROC (Receiver Operating Characteristic) curve with the diagnostic trial result of the selected diagnostic model Md (step S202).
Further, the diagnostic performance determination unit 114 calculates an AUC (Area Under the Curve) area from the plotted ROC curve (step S203). The ROC curve and AUC area will be described later.
Then, the diagnostic performance determination unit 114 determines whether or not the processing of Step S202 and Step S203 has been completed for the diagnostic model Md corresponding to all the diagnostic trial results (S204).

As a result of step S204, when the processes of step S202 and step S203 have not been completed for the diagnostic model Md corresponding to all the diagnostic trial results (S204 → No), the diagnostic performance determination unit 114 returns the process to step S201.
As a result of step S204, when the processes of step S202 and step S203 have been completed for the diagnostic model Md corresponding to all the diagnostic trial results (S204 → Yes), the diagnostic performance determination unit 114 is calculated in step S203. The diagnostic model Md corresponding to the diagnostic trial result with the largest AUC area is selected (S205). The diagnostic model Md selected in step S205 is the diagnostic model Md with the best diagnostic performance.

In the ROC curve, the vertical axis is the diagnosis success rate (= the rate at which the abnormality of the device 2 is correctly reported), and the horizontal axis is the false alarm rate (= the rate at which the device 2 is normally reported). It is the graph which plotted the diagnostic performance at the time of changing the operating condition of Md. The AUC area is the area under the ROC curve and indicates diagnostic performance.

In this embodiment, the brute force method is used as the diagnostic trial method and AUC maximization is used as the diagnostic performance evaluation method, but this is not restrictive. For example, target values for the false alarm rate and the false alarm rate may be defined as the diagnostic performance evaluation method, and the annealing method or the like may be used as the diagnostic trial method.

FIG. 10 is a diagram illustrating a result of the diagnostic performance determination process according to the first embodiment. 10, the same components as those in FIG. 7 are denoted by the same reference numerals and description thereof is omitted.
As a result of the diagnostic performance determination unit 114 performing the diagnostic performance determination process shown in FIG. 9, the diagnostic processing unit Sw3 (Sw), the common condition Pc3 (Pc) ), A combination of individual conditions Pm3 (Pm) is selected. The common condition Pc3 has an input condition Pi1 (Pi), a learning condition Pt2 (Pt), and a diagnosis condition Pd2 (Pd). The input condition Pi1, the learning condition Pt2, and the diagnostic condition Pd2 are managed in association with the input unit Si2 (Si), the learning unit St2 (St), and the diagnostic unit Sd1 (Sd), which are components of the diagnostic processing unit Sw3. . The individual condition Pm3 is managed in association with the learning unit St2 and the diagnostic unit Sd1 (see FIG. 2).

(Diagnostic job template)
FIG. 11 is a diagram illustrating an example of a diagnostic job template and a diagnostic job according to the first embodiment. FIG. 11A is a diagram illustrating an example of a diagnostic job template according to the first embodiment. () Is a diagram showing an example of a diagnostic job according to the first embodiment.
First, a diagnostic job template according to the first embodiment will be described with reference to FIG.
The diagnosis job template Mda is a template of the diagnosis job 122 (FIG. 1) corresponding to each of the devices 2 and 1: 1, and learning is performed from the diagnosis model Md selected by the diagnosis performance determination processing of FIG. The part St and the learning condition Pt are excluded. Therefore, the diagnostic job template Mda is composed of the input unit Si and the input condition Pi that is its operating condition, and the diagnostic unit Sd and the diagnostic condition Pd that is its operating condition.

The diagnostic job template Mda shown in FIG. 11A is judged to have the best diagnostic performance in the cube 400 as shown in FIG. 10, and the selected diagnostic processing unit Sw3 (Sw), common condition Pc3 (Pc ), The learning unit St2 (St) and the learning condition Pt2 (Pt) are deleted. The diagnosis processing unit Sw3 from which the learning unit St3 has been deleted is referred to as a diagnosis processing unit Sw3a (Swa), and the common condition Pc3 from which the learning condition Pt3 has been deleted is referred to as a common condition Pc3a (Pca).

(Diagnostic job)
Next, a diagnostic job according to the first embodiment will be described with reference to FIG.
The diagnosis job 122 is obtained by adding a code book CB3 (CB) as an individual condition Pm to the diagnosis job template Mda of FIG.
Note that the diagnosis job template Mda shown in FIG. 11A is data associated with the “model” of the device 2 and cannot perform actual diagnosis because there is no codebook CB. On the other hand, since the diagnosis job 122 includes the code book CB3 (CB) corresponding to the device 2, it can be executed and is associated with each device 2.

The diagnostic job 122a (122) shown in FIG. 11B is obtained by applying the individual condition Pm3 (Pm), that is, the code book CB3 (CB) to the diagnostic job template Mda shown in FIG. It is. Here, assuming that the code book CB3 corresponds to the first machine of the device 2, the diagnostic job 122a is the diagnostic job 122 corresponding to the first machine.

(Diagnostic job generation processing)
FIG. 12 is a flowchart illustrating a procedure of diagnostic job generation processing according to the first embodiment. Reference is made to FIG. 3 as appropriate.
First, the diagnostic job generation unit 115 selects the diagnostic model Md with the best diagnostic performance selected in step S205 in the diagnostic performance determination process of FIG. 9 (S301), and the diagnostic processing unit Sw of the selected diagnostic model Md. Select.
Next, the diagnosis job generation unit 115 deletes the learning unit St from the selected diagnosis processing unit Sw (S302). Here, the diagnosis job generation unit 115 deletes the program corresponding to the learning unit St from the program in the diagnosis processing unit Sw, or deactivates the execution of the job of the learning unit St.

Then, the diagnostic job generation unit 115 acquires the common condition Pc (specifically, the input condition Pi and the diagnostic condition Pd) included in the diagnostic model Md having the best diagnostic performance (S303).

Furthermore, the diagnostic job generation unit 115 has the type to be processed and acquires input data (sensor data) during normal operation from each unprocessed device 2 (S311).
Next, the diagnostic job generation unit 115 performs learning processing by applying the input data acquired in step S311 to the learning unit St (S312). In step S312, the same learning process as the learning process performed by the learning unit St deleted in step S302 is performed using the input data obtained from each device 2. Note that the learning result in step S312 is stored in the code book CB associated with the device 2 to be processed.

Subsequently, the diagnostic job generation unit 115 acquires the code book CB generated as a result of the learning process by the learning unit St in step S312 and the diagnostic processing unit Sw from which the learning unit St has been deleted in step S302, in step S303. This is applied to the diagnostic job template Mda configured with the common condition Pc (S313). As a result of step S313, a diagnostic job 122 for the device 2 to be processed is generated.
Thereafter, the diagnostic job generation unit 115 holds the generated diagnostic job 122 in the storage device 120 (step S314).

Thereafter, the diagnostic job generation unit 115 determines whether or not the processing from step S311 to step S314 has been performed (completed) for all the devices 2 having the model to be processed (S315).
As a result of step S315, if the processing from step S311 to step S314 has not been completed for all the devices 2 having the type to be processed (S315 → No), the diagnostic job generation unit 115 proceeds to step S311. To return.
As a result of step S315, when the processing from step S311 to step S314 is performed for all the devices 2 having the type to be processed (S315 → Yes), the diagnostic job generation unit 115 ends the diagnostic job generation processing. To do.

(Diagnostic job)
FIG. 13 is a diagram illustrating an example of a diagnostic job according to the first embodiment, in which (a) illustrates an example of a diagnostic job for the second machine, (b) illustrates an example of a diagnostic job for the third machine, and (c) ) Shows an example of a diagnostic job for Unit N. In FIG. 13, the same components as those in FIG. 11A are denoted by the same reference numerals and description thereof is omitted.
Since the diagnostic job generation unit 115 generates one diagnostic job 122 for each device 2, a diagnostic job 122 is generated for each device number of the device 2 (here, “No. 1” to “N”). . However, since the diagnosis job 122a for the first machine is shown in FIG. 11B, the diagnosis jobs 122b to 122d (122) for the second, third, and N machines are shown in FIG. The diagnosis jobs 122 for the No. 4 to No. 1 machines are the same as the diagnosis jobs 122b to 122d for the No. 2, No. 3, and No. N machines, and thus the description thereof is omitted here. The diagnosis processing unit Sw3a and the common condition Pc3a of the diagnosis job 122 for each unit are all the same as in FIG. 11B, and only the code book CB is different. The code book CB is generated by performing learning processing based on the normal input data obtained from each device 2 in step S311 in FIG. 12 (S312 in FIG. 12). ) To (c), the diagnostic job 122 of each device 2 is associated with the 1: 1.

That is, the diagnostic job 122b shown in FIG. 13A is obtained by applying the codebook CB32 (CB) for the second machine to the diagnostic job template Mda of FIG. 11A.
A diagnostic job 122c shown in FIG. 13B is obtained by applying the codebook CB33 (CB) for Unit 3 to the diagnostic job template Mda of FIG. 11A.
Similarly, the diagnostic job 122d shown in FIG. 13C is obtained by applying the code book CB34 (CB) for the No. N machine to the diagnostic job template Mda of FIG. 11A.

(Diagnostic job list screen)
FIG. 14 is a diagram illustrating an example of a diagnostic job list screen according to the first embodiment.
The diagnostic job list screen 500 is a screen displayed on the display device 350.
The diagnosis job list screen 500 is a list including a job ID, a target device, a diagnosis model, device information, and diagnosis performance.
The job ID is an ID assigned 1: 1 to the diagnostic job 122 generated by the diagnostic job generation unit 115 in association with each device 2. The diagnostic job generation unit 115 assigns a job ID when the diagnostic job 122 is generated.

In the target device field, the device name of the device 2 that is a diagnosis target by the diagnosis job 122 and has the job ID is displayed. That is, the device name displayed in the field of the target device 2 is the device 2 associated with the code book CB to which the diagnostic job 122 is applied.

In the diagnostic model field, the name of the model (model name) that is the target of the diagnostic model Md is displayed as information associated with the device model. In this way, the user can determine which type of diagnostic model Md each diagnostic job 122 uses.
In the device information field, a device number is displayed as information associated with the device 2.
In the diagnostic performance field, an upper limit and a lower limit are displayed for each type of the false alarm rate and the false alarm rate (execution diagnostic performance) acquired as a result of executing the diagnostic job 122 generated for the devices 2 in the same model. The The diagnostic performance is calculated from data obtained by actually executing the diagnostic job 122 for each device 2 after the diagnostic job 122 is generated. The execution diagnosis performance may be a result of causing the device 2 to execute the diagnosis job 122 on a trial basis.

By displaying the diagnostic job list screen 500 as shown in FIG. 14, the user can easily confirm which model the generated diagnostic model Md corresponds to. Furthermore, by displaying the diagnostic job list screen 500 as shown in FIG. 14, the user can determine which diagnostic model Md is the basis of the generated diagnostic job 122 and whether the diagnostic performance of the diagnostic model Md is sufficient. Can be easily confirmed. Furthermore, by displaying the diagnostic job list screen 500 as shown in FIG. 14, the user can easily confirm the correspondence relationship between the device 2 and the diagnostic model Md.
14 are the device 2 selected by the user, the device 2 in which an abnormality is detected as a result of executing the diagnosis removal unit 122, and the like.

(Summary of the first embodiment)
In the present embodiment, first, the diagnostic trial unit 113 performs a diagnostic trial on a plurality of diagnostic models Md and selects a diagnostic model Md with the best diagnostic performance. Then, the diagnostic job generation unit 115 generates the diagnostic job 122 by applying the individual condition Pm corresponding to each device 2 to the selected best diagnostic model Md. For example, in the technique described in Patent Literature 1, when a new diagnosis is performed on the device 2, a statistical model of the existing device 2 is performed one by one to generate a diagnosis model. On the other hand, in this embodiment, an optimal diagnostic model can be efficiently generated by performing a diagnostic trial on a plurality of diagnostic models Md and selecting a diagnostic model Md having the best diagnostic performance. Furthermore, a diagnostic job 122 can be generated by generating a code book CB for the device 2 that has been newly diagnosed and applying the generated code book CB to the diagnostic job template Mda that has been generated in advance. As a result, the number of steps for generating the diagnostic job 122 can be reduced.

The diagnostic model generation unit 112 generates a plurality of diagnosis models Md by changing each of the input unit Si, the learning unit St, and the diagnosis unit Sd (a plurality of processing units) in the diagnosis processing unit Sw, thereby generating a plurality of diagnosis models Md. The generation of the diagnostic model Md can be performed efficiently.
In the present embodiment, the diagnostic job template Mda is generated by deleting the learning unit St and the learning condition Pt from the diagnostic model Md having the best diagnostic performance. In this way, the diagnostic job template Mda and the diagnostic job 122 can be easily generated.

Further, in the present embodiment, the content of the individual condition Pm is the code book CB that is the result of the learning process. By doing in this way, the diagnostic job production | generation apparatus 1 produces | generates the code book CB only by defining the content of the learning part St and the learning condition Pt. That is, the user's labor can be greatly reduced, and efficient generation of a diagnostic job can be realized.
Further, the diagnostic job generation system 10 according to the present embodiment can share the diagnostic model Md between the devices 2 having the same type.

[Second Embodiment]
FIG. 15 is a diagram illustrating a diagnostic model according to the second embodiment. Further, in FIG. 15, the same components as those in FIG. Furthermore, in the second embodiment, since the configuration other than the configuration shown in FIG. 15 is the same as that of the first embodiment, illustration and description thereof are omitted.
The diagnostic model Mdz shown in FIG. 15 is obtained by adding a preprocessing unit Sr before the learning unit St and adding a postprocessing unit Sp after the diagnostic unit Sd to the diagnostic model Md in the first embodiment. .

The diagnosis model Mdz in FIG. 15 includes a diagnosis processing unit Swz, a common condition Pcz, and an individual condition Pm, as in the first embodiment.
The diagnostic processing unit Swz includes an input unit Si, a preprocessing unit Sr, a learning unit St, a diagnostic unit Sd, and a post-processing unit Sp.
The common condition Pcz includes an input condition Pi, a preprocessing condition Pr, a learning condition Pt, a diagnosis condition Pd, and a post-processing condition Pp. The input unit Si, the pre-processing unit Sr, the learning unit St, the diagnostic unit Sd, and the post-processing, respectively. The operating conditions of the part Sp are determined.
The individual condition Pm is the same as in the first embodiment.

As shown in FIG. 15, in the diagnosis model Mdz, the preprocessing unit Sr is provided in the preceding stage of the learning unit St, for example, so that the following can be performed. That is, for example, when diagnosing a structure such as a bridge, the vibration sensor data as input data is not used as it is, but the power spectrum is calculated by Fourier transforming the vibration data as the pre-processing unit Sr. And the power spectrum which is a pre-processing result is input into the learning part St and the diagnostic part Sd, and it processes based on the pre-processing result in which the learning part St and the diagnostic part Sd were input.
Alternatively, the preprocessing unit Sr can apply the moving average or the like as the preprocessing so that the learning processing by the learning unit St can be performed after removing noise from the input data.

Also, as shown in FIG. 15, by providing the post-processing unit Sp after the diagnostic processing, for example, instead of directly using the result of the diagnostic unit Sd, it is possible to make another determination. This is due to the following reasons. In other words, if the diagnosis unit Sd returns a result of device abnormality, immediate reporting will increase the number of false reports. Therefore, for example, the post-processing unit Sp performs a monitoring process for a certain period, so that a determination is made such that the alarm is issued only when it is regarded as abnormal for a certain period of time.

In the second embodiment, in the generated diagnostic job template Mda (FIG. 11) and diagnostic job 122, a pre-processing unit Sr is inserted between the input unit Si and the diagnostic unit Sd, and the subsequent stage of the diagnostic unit Sd. Is followed by the post-processing unit Sp. Further, the diagnosis job template Mda (FIG. 11) and the diagnosis job 122 generated in the second embodiment include a preprocessing condition Pr and a post-processing associated with the preprocessing unit Sr as the common condition Pca (FIG. 11). A post-processing condition Pp associated with the part Sp is provided.
Note that both the pre-processing unit Sr and the post-processing unit Sp are not necessarily provided, and either one may be provided. The preprocessing condition Pr and the postprocessing condition Pp can be omitted depending on the contents of the preprocessing unit Sr and the postprocessing unit Sp. A plurality of at least one of the preprocessing unit Sr and the postprocessing unit Sp may be provided.

According to the second embodiment, when learning using input data as it is is impossible or when it is inconvenient to use the diagnostic trial result as it is, appropriate learning processing and output of the diagnostic trial result are performed. Can do.

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment has been described in detail for easy understanding of the present invention, and is not necessarily limited to having all the configurations described. In addition, a part of the configuration of a certain embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of a certain embodiment. In addition, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

Further, each of the above-described configurations, functions, units 111 to 115, 311, 312, and storage devices 120 and 320 may be realized by hardware by designing a part or all of them, for example, by an integrated circuit. Good. Further, as shown in FIG. 3, the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by a processor such as a CPU. In addition to storing information such as programs, tables, and files for realizing each function in the HD, a memory, a recording device such as an SSD (Solid State Drive), an IC (Integrated Circuit) card, an SD (Secure It can be stored in a recording medium such as a Digital) card or DVD (Digital Versatile Disc).
In each embodiment, control lines and information lines are those that are considered necessary for explanation, and not all control lines and information lines are necessarily shown on the product. In practice, it can be considered that almost all configurations are connected to each other.

DESCRIPTION OF SYMBOLS 1 Diagnosis job generation apparatus 2 Apparatus 10 Diagnosis job generation system 111 Processing part 112 Diagnosis model generation part 113 Diagnosis trial part (extraction part)
114 Diagnostic performance determination unit (extraction unit)
115 Diagnostic Job Generation Unit 121 History Information 122, 122a, 122b, 122c, 122d Diagnostic Job 312 Display Processing Unit 350 Display Device (Display Unit)
CB, CB1, CB2, CB3, CB32, CB33, CB34 Codebook H, H1, H2, H3, H11, H12, H13, H21, H22, H23 History Md, Mdz Diagnostic model Mda Diagnostic job template Pc, Pca, Pcz , Pc1, Pc2, Pc3, Pc3a, Pc4 Common conditions Pd, Pd1, Pd2 Diagnostic conditions Pi, Pi1 Input conditions Pm, Pm1, Pm2, Pm3, Pm4, Pm32, Pm33, Pm34 Individual conditions Pp Post-processing conditions Pr Pre-processing conditions Pt , Pt1, Pt2 Learning condition Sd, Sd1 Diagnostic unit Si, Si1, Si2 Input unit Sr Pre-processing unit Sp Post-processing unit St, St1, St2 Learning unit Sw, Swa, Swz, Sw1, Sw2, Sw3, Sw3a Diagnostic processing unit

Claims (17)

A diagnostic model generation unit that generates a plurality of common diagnostic models applicable to devices belonging to the same group;
An extraction unit that performs a diagnosis for evaluating the performance of the plurality of generated diagnosis models for each diagnosis model, and extracts a predetermined diagnosis model from the plurality of generated diagnosis models based on the diagnosis result;
A diagnostic job generation unit that generates a diagnostic job for each device by applying an individual condition that is a condition corresponding to each of the devices to the extracted diagnostic model;
A diagnostic job generation system comprising: The diagnostic model has a plurality of processing units,
The diagnostic model generation unit
The diagnostic job generation system according to claim 1, wherein a plurality of the diagnostic models are generated by changing a combination of the plurality of processing units. The diagnostic model includes a diagnostic processing unit that diagnoses the device, and a common condition that is an operation condition applied to the diagnostic processing unit,
The diagnostic processing unit includes a learning unit that performs a learning process based on input data, and a diagnostic unit that diagnoses the device using a learning result that is a result of the learning unit,
The common condition includes a learning condition that is an operation condition of the learning unit and a diagnosis condition that is an operation condition when diagnosing the device.
The diagnostic model generation unit generates a plurality of the diagnostic processing units by generating a combination of the different learning unit and the diagnostic unit, and combines the different learning conditions and the diagnostic conditions. By generating a plurality of the common conditions,
The extraction unit combines the generated diagnostic processing unit and the common condition, and then applies the individual condition to execute the diagnostic processing unit. As a result of the execution, the diagnostic performance is the best. A combination of a diagnostic processing unit and the common condition is extracted as the diagnostic model with the best diagnostic performance,
The diagnostic job generation unit deletes the learning unit and the learning condition included in the diagnostic model from the diagnostic model extracted by the extraction unit, and removes the learning unit from each of the diagnostic processing units. The diagnostic job generation system according to claim 1, wherein a diagnostic job for each device is generated by applying an individual condition corresponding to the device. The diagnostic job generation system according to claim 3, wherein the individual condition is a result of the diagnostic job generation unit performing a learning process performed by the deleted learning unit. The diagnostic model is
The diagnostic job generation system according to claim 3, comprising at least one of a pre-processing unit that performs pre-processing before the learning unit and a post-processing unit that performs post-processing after the diagnostic unit. A diagnostic job generation system that generates a diagnostic job for diagnosing equipment
Generate multiple common diagnostic models applicable to devices belonging to the same group,
A diagnosis for evaluating the performance of the plurality of generated diagnosis models is performed for each diagnosis model, and based on the diagnosis result, a predetermined diagnosis model is extracted from the plurality of generated diagnosis models,
A diagnostic job generation method for generating a diagnostic job for each device by applying an individual condition, which is a condition corresponding to each of the devices, to the extracted diagnostic model. The diagnostic model has a plurality of processing units,
The diagnostic job generation system includes:
The diagnostic job generation method according to claim 6, wherein a plurality of the diagnostic models are generated by changing a combination of the plurality of processing units. The diagnostic model includes a diagnostic processing unit that diagnoses the device, and a common condition that is an operation condition applied to the diagnostic processing unit,
The diagnostic processing unit includes a learning unit that performs a learning process based on input data, and a diagnostic unit that diagnoses the device using a learning result that is a result of the learning unit,
The common condition includes a learning condition that is an operation condition of the learning unit and a diagnosis condition that is an operation condition when diagnosing the device.
The diagnostic job generation system includes:
A plurality of the diagnostic processing units are generated by generating a combination of the different learning unit and the diagnostic unit, and a common combination is generated by generating a combination of the different learning condition and the diagnostic condition. Generate multiple conditions,
After combining the generated diagnostic processing unit and the common condition, applying the individual condition, causing the diagnostic processing unit to execute, and as a result of the execution, the diagnostic processing unit having the best diagnostic performance, Extracting the combination with the common condition as the diagnostic model with the best diagnostic performance;
By deleting the learning unit and the learning condition included in the diagnostic model extracted from the extracted diagnostic model, and applying the individual condition corresponding to each device to the diagnostic processing unit from which the learning unit has been deleted The diagnostic job generation method according to claim 6, wherein a diagnostic job for each device is generated. The diagnostic job generation method according to claim 8, wherein the individual condition is a result of a learning process performed by the deleted learning unit performed by the diagnostic job generation system. The diagnostic model is
The diagnostic job generation method according to claim 8, comprising at least one of a pre-processing unit that performs pre-processing before the learning unit and a post-processing unit that performs post-processing after the diagnostic unit. A diagnostic job generation system that generates a diagnostic job for diagnosing equipment
Generate multiple common diagnostic models applicable to devices belonging to the same group,
A diagnosis for evaluating the performance of the plurality of generated diagnosis models is performed for each diagnosis model, and based on the diagnosis result, a predetermined diagnosis model is extracted from the plurality of generated diagnosis models,
A diagnostic job for each device is generated by applying an individual condition that is a condition corresponding to each device to the extracted diagnostic model,
A diagnostic job generation and display method, wherein a relationship between the diagnostic model and a model in the device is displayed on a display unit. The diagnostic job generation system includes:
The execution diagnostic performance when the generated diagnostic job is applied to a device is displayed on the display unit in association with information of a diagnostic model from which the diagnostic job is generated. The diagnostic job generation display method according to claim 11. The diagnostic job generation system includes:
The diagnostic job generation and display method according to claim 11, wherein a relationship between the diagnostic job and the device is displayed on the display unit. The diagnostic model has a plurality of processing units,
The diagnostic job generation system includes:
The diagnostic job generation and display method according to claim 11, wherein a plurality of the diagnostic models are generated by changing a combination of the plurality of processing units. The diagnostic model includes a diagnostic processing unit that diagnoses the device, and a common condition that is an operation condition applied to the diagnostic processing unit,
The diagnostic processing unit includes a learning unit that performs a learning process based on input data, and a diagnostic unit that diagnoses the device using a learning result that is a result of the learning unit,
The common condition includes a learning condition that is an operation condition of the learning unit and a diagnosis condition that is an operation condition when diagnosing the device.
The diagnostic job generation system includes:
A plurality of the diagnostic processing units are generated by generating a combination of the different learning unit and the diagnostic unit, and a common combination is generated by generating a combination of the different learning condition and the diagnostic condition. Generate multiple conditions,
After combining the generated diagnostic processing unit and the common condition, applying the individual condition, causing the diagnostic processing unit to execute, and as a result of the execution, the diagnostic processing unit having the best diagnostic performance, Extracting the combination with the common condition as the diagnostic model with the best diagnostic performance;
By deleting the learning unit and the learning condition included in the diagnostic model extracted from the extracted diagnostic model, and applying the individual condition corresponding to each device to the diagnostic processing unit from which the learning unit has been deleted The diagnostic job generation display method according to claim 11, further comprising: generating a diagnostic job for each device. The diagnostic job generation and display method according to claim 15, wherein the individual condition is a result of the learning process performed by the deleted learning unit performed by the diagnostic job generation system. The diagnostic model is
16. The diagnostic job generation and display method according to claim 15, comprising at least one of a pre-processing unit that performs pre-processing before the learning unit and a post-processing unit that performs post-processing after the diagnostic unit. .

Images