JP5115025B2 - Fault diagnosis system and fault diagnosis program - Google Patents

Description

  The present invention relates to a failure diagnosis system and a failure diagnosis program.

  In recent years, electronic devices such as personal computers, copiers, scanners, printers, and multi-functional devices in which these are combined have been improved in terms of performance and functions. It has been attached in the form of PWBA (referred to as Printed Wiring Board Assembly). In order to realize a series of functions, PWBAs are connected to each other through wiring such as cables in various forms, and exchange of signals is performed. An environment in which such a device on which PWBA is mounted is used mainly in an office or a house, but it may be used in other severe environments and is very diverse.

  For this reason, various abnormalities occur in each situation, and much effort is required to repair these abnormalities and failures. In addition, the frequency of occurrence of any abnormality is not low, and when an abnormality occurs, an immediate response is required from the viewpoint of man-hours and costs.

  By the way, with the aim of providing proactive online diagnostics in manageable networks, update at least one of the probability of malfunction based on alarms and causal networks and present alarm diagnostics in response to the updated probabilities It has been proposed (see Patent Document 1).

  It is also proposed that when a failure occurs in a device, the diagnosis engine refers to the failure diagnosis database based on the device identification information and the failure mode, and analyzes failures other than secular changes based on operation results. (See Patent Document 2).

Furthermore, multiple computers such as web systems are related to each other, monitoring active systems and detecting abnormalities online, and detecting transitions to abnormal states by monitoring feature vectors using probability models Has been proposed (see Patent Document 3).
JP 2003-032253 A JP 2005-182465 A Japanese Patent Application Laid-Open No. 2005-216066

  However, in Patent Document 1 to Patent Document 3, the relationship between evidence information that should originally be included in the diagnostic model cannot be expressed completely, and the flexibility of the diagnostic system (for example, saving of trouble handling time and man-hours) In other words, there is a problem that the accuracy of diagnosis cannot be improved because the burden on the customer is reduced and the failure diagnosis is easy.

  In consideration of the above facts, the present invention expresses a relationship between pieces of evidence information that should originally be included in a diagnostic model, gives the flexibility of the diagnostic system, and improves the diagnostic accuracy. The purpose is to obtain a diagnostic program.

  According to the first aspect of the present invention, a node generation unit that generates a node by modeling at least one of hardware or software, which is a plurality of original resources existing as functionally necessary resources, and a failure diagnosis target The plurality of nodes are connected in association with each other, a failure determination unit that determines a failure based on history information output from a system expressed by the plurality of nodes, a failure determination unit that is acquired from the determination result by the failure determination unit, Unlike at least one of hardware and software, which is a resource, modeling evidence information resources in which no entity exists, virtual node generation means for generating virtual nodes, and the plurality of virtual nodes are mutually linked and modeled. A collaborative node generating means for generating a collaborative node, and evidence information of the virtual node by the collaborative node. An inferred analogy means, a plurality of virtual nodes whose evidence information has been clarified by analogy by the analogy means, and a node that is out of order by changing at least the pass / fail judgment balance indicated by both of the cooperation nodes And a specifying means for specifying.

  According to the first aspect of the present invention, a node is generated by modeling at least one of hardware or software, which is a plurality of original resources existing as functionally necessary resources, by the node generation unit. Note that the raw resource includes one or both of hardware and software, and particularly includes the case where the software is involved in a plurality of hardware. Further, the failure determination means connects the plurality of nodes to be subjected to failure diagnosis in association with each other, and determines failure based on the history information output from the plurality of nodes. Further, a virtual node is generated by modeling the evidence information resource which is acquired from the determination result by the failure determination unit by the virtual node generation unit and which is different from the original resource and does not exist. And a cooperation node production | generation means mutually cooperates and models the said some virtual node, and produces | generates a cooperation node. Therefore, the analogy means analogizes the evidence information of the virtual node by the cooperation node, and the specifying means indicates both the plurality of virtual nodes whose evidence information has been clarified by analogy by the analogy means and the cooperation node. By changing at least the pass / fail judgment balance, the failed node is identified.

  Therefore, it expresses the relationship between the evidence information that should be included in the diagnosis model, and the flexibility of the diagnosis system, for example, saving time and man-hours for failure, reducing the burden on customers, and ease of failure diagnosis, etc. Can improve diagnostic accuracy.

  According to a second aspect of the present invention, the failure diagnosis system according to the first aspect is executed by a failure diagnosis program based on a Bayesian network, and the plurality of virtual nodes and the cooperation node fail when the failure diagnosis program is executed. It becomes the unit which specifies.

  According to the second aspect of the present invention, the linkage node is generated by the failure diagnosis program based on the Bayesian network, so that the evidence information is analogized and the relationship between the evidence information that should be included in the diagnosis model is expressed. Thus, the flexibility of the failure diagnosis system can be provided, and the diagnosis accuracy of failure diagnosis can be improved.

The invention according to claim 3, the node generating means for a computer, hardware or modeling at least one of the software is a plurality of original resources present as functionally necessary resources, to generate a node, fault diagnosis The plurality of target nodes are connected in association with each other, a failure determination unit that determines a failure based on history information output from a system represented by the plurality of nodes, and obtained from a determination result by the failure determination unit Unlike at least one of the hardware and software that are the original resources, virtual node generation means for generating a virtual node by modeling an evidence information resource having no entity, and the plurality of virtual nodes are mutually linked. , by modeling the virtual, and the coordination node generating means for generating a coordination node, by the coordination node Bruno And analogy means for estimating an evidence information de, and the plurality of virtual nodes evidence information by analogy by the analogy means revealed by changing respective determination balance of at least quality both indicate the coordination node, It is a fault diagnosis program for functioning as a specifying unit that specifies a faulty node.

According to the invention described in claim 3, the computer is caused to function as each means described in claim 3. That is, the node generation unit generates a node by modeling at least one of hardware or software, which is a plurality of original resources existing as functionally necessary resources. Note that the raw resource includes one or both of hardware and software, and particularly includes the case where the software is involved in a plurality of hardware. Further, the failure determination means connects the plurality of nodes to be subjected to failure diagnosis in association with each other, and determines failure based on the history information output from the plurality of nodes. Further, a virtual node is generated by modeling the evidence information resource which is acquired from the determination result by the failure determination unit by the virtual node generation unit and which is different from the original resource and does not exist. And a cooperation node production | generation means mutually cooperates and models the said some virtual node, and produces | generates a cooperation node. Therefore, the analogy means analogizes the evidence information of the virtual node by the cooperation node, and the specifying means indicates both the plurality of virtual nodes whose evidence information has been clarified by analogy by the analogy means and the cooperation node. By changing at least the pass / fail judgment balance, the failed node is identified.

  Therefore, it expresses the relationship between the evidence information that should be included in the diagnosis model, and the flexibility of the diagnosis system, for example, saving time and man-hours for failure, reducing the burden on customers, and ease of failure diagnosis, etc. Can improve diagnostic accuracy.

  According to a fourth aspect of the present invention, the failure diagnosis program according to the third aspect is executed by a failure diagnosis program based on a Bayesian network, and the plurality of virtual nodes and the cooperation node fail when the failure diagnosis program is executed. It becomes the unit which specifies.

  According to the fourth aspect of the present invention, the linkage node is generated by the failure diagnosis program based on the Bayesian network, whereby the evidence information is analogized and the relationship between the evidence information that should be included in the diagnosis model is expressed. Thus, the flexibility of the failure diagnosis program can be provided, and the diagnosis accuracy of failure diagnosis can be improved.

  In the present invention, what is called a “node” is a node that means an entity, unlike the “cooperative node” and the “virtual node”. Therefore, hereinafter, a “node” having this entity is referred to as an “entity node” in order to distinguish it from the “cooperation node” and the “virtual node”.

  As described above, according to the present invention, a failure diagnosis system that expresses the relationship between evidence information that should originally be included in the diagnosis model, has the flexibility of the diagnosis system, and can improve the diagnosis accuracy. And the effect that a failure diagnosis program can be obtained is acquired.

(First embodiment)
FIG. 1 is a schematic diagram showing an overall configuration 100 of a network for applying failure diagnosis according to the first embodiment and devices connected thereto.

  The overall configuration 100 includes a desktop personal computer 110, a notebook personal computer 120, and devices 130a and 130b (for example, image forming apparatuses such as printers, copiers, and multifunction devices).

  The desktop personal computer 110 is connected to the device 130a and the device 130b via the first communication line 140. The notebook personal computer 120 is also connected to the device 130 a via the second communication line 150.

  In FIG. 1, the devices 130a and 130b are diagnosed (particularly, failure diagnosis) by the desktop personal computer 110 and the notebook personal computer 120. Alternatively, the devices 130a and 130b are diagnosed via the first communication line 140.

  The first communication line 140 and the second communication line 150 are a line network (for example, a local area network (LAN) or a wide area network (WAN)) such as a network (for example, a wide area communication network). Internet line network, etc.) may also be used. Further, the first communication line 140 and the second communication line 150 may be a wired or wireless communication network.

  FIG. 2 is a simplified block diagram 200 of the configuration of the PWBA that controls the devices 130a and 130b of FIG.

  The simplified block diagram 200 is composed of PWBA (A) 202, PWBA (B) 204, PWBA (C) 206, PWBA (D) 208, and PWBA (E) 210 (collectively referred to as PWBA). Yes. Since these PWBAs are physical entities, they are hereinafter referred to as “original resources” in the first embodiment, the second embodiment, and the third embodiment. Further, for example, the “raw resource” includes one or both of hardware and software, and particularly includes a case where the software is related to a plurality of hardware.

  PWBA (A) 202 is connected toward PWBA (C) 206, PWBA (E) 210 is connected toward PWBA (C) 206 and PWBA (D) 208, and PWBA (C) 206 is connected to PWBA (C) 206. D) connected to 208 and PWBA (D) 208 is connected to PWBA (B) 204.

  Note that the above connection relationship indicates the connection relationship indicated by the dotted line in FIG. 2. Normally, PWBA (A) 202 is connected to PWBA (C) 206, and PWBA (C) 206 is connected to PWBA (D). 208 and PWBA (E) 210, and PWBA (D) 208 is connected to PWBA (E) 210 and PWBA (B) 204.

  As shown in the connection relationship of the arrows shown by the dotted lines in FIG. 2 above, since the PWBAs are connected, an information code (also called output information) that is generally output as failure information (also called evidence information) ) FailCode (1) 212 and FailCode (2) 214 are output from the PWBA (B) 204.

  The FailCode (1) 212 and the FailCode (2) 214 are user interfaces (User Interfaces) in a simple manner in which the failure status is in a code format (for example, a file format in which data such as CSV format is separated by commas). (Hereinafter referred to as “UI”), and is often mounted in advance in the device in order to notify the user or diagnosis support person of the condition. Separately from this, an inspection program for narrowing down a part to be diagnosed for failure, a log (log) that can know the state of only a specific function although a plurality of control areas are involved, and the like are also failure information. It contributes to diagnosis by treating it as evidence information.

  For example, Log is a usage status of a device such as a computer or a record of data communication, and is also called history information. Log also includes a general log, a local inspection log that is a result of inspecting a local part of a device by a program, a software (Software) inspection log that is a result of inspecting software operating on the device by a program, and the like. . If there is a result to be output, any form (for example, the phenomenon of freeze or hang-up) can be handled as evidence information.

  FIG. 3 is a model 300 of a Bayesian network when a failure point is inferred by a Bayesian network that is a causal network based on FIG. 2, and the last two digits of each symbol correspond to each PWBA of FIG. Yes.

  A Bayesian network represents a problem area where causal relationships are complex. Therefore, a causal relationship between a plurality of variables (nodes) associated with a conditional probability distribution is represented by a directed edge (connection between nodes). The arcs shown (arrows in the figure) are connected in order, expressed as a network having a graph structure, and the dependency between nodes is represented by a directed graph. That is, a Bayesian network models a problem area using probability theory. Arrows called arcs indicate causal relationships and are linked from cause to result in the direction of the arrow. Each node includes a conditional probability table that represents the probability of a state occurring in a child node (arrow tip side) when the parent node (base arrow side) has a predetermined probability.

  The model 300 includes a PWBA (A) node 302, a PWBA (B) node 304, a PWBA (C) node 306, a PWBA (D) node 308, and a PWBA node (E) 310 (collectively referred to as PWBA nodes). It consists of The PWBA node is an actual node.

  Further, the model 300 includes a FailCode (1) node 312 and a FailCode (2) node 314, an evidence information (L1) node 316, an evidence information (L2) node 318, an evidence information (L3) node 320, and an evidence information linkage (L1). ) Node 322 and evidence information linkage (L2) node 324 are also included.

  The PWBA (A) node 302 is connected by an arc toward the PWBA (C) node 306 and the evidence information (L1) node 316, and the PWBA (E) node 310 is connected to the PWBA (C) node 306, the PWBA (D) node 308, And the evidence information (L3) is connected to the node 320 by an arc. The PWBA (C) node 306 is connected to the PWBA (D) node 308, the FailCode (1) node 312, the evidence information (L1) node 316, and the evidence information (L2) node 318 by an arc. Further, the PWBA (D) node 308 is connected by an arc toward the PWBA (B) node 304, the evidence information (L2) node 318, and the evidence information (L3) node 320, and the PWBA (B) node 304 is connected to the FailCode (1). ) Node 312, FailCode (2) node 314, evidence information (L 1) node 316, and evidence information (L 2) node 318. Also, the FailCode (1) node 312 is connected by an arc toward the evidence information cooperation (L1) node 322 and the evidence information cooperation (L2) node 324, and the FailCode (2) node 314 is connected to the evidence information cooperation (L2) node 324. Connected by arc toward. Further, the evidence information (L1) node 316 is connected by an arc toward the evidence information cooperation (L2) node 324, and the evidence information (L2) node 318 and the evidence information (L3) node 320 are connected to the evidence information cooperation (L1) node 322. Connected by arc towards.

  For example, in the connection relationship, according to the Bayesian network, the PWBA (A) node 302 is a parent node, and the PWBA (C) node 306 and the evidence information (L1) node 316 are directed to two child nodes. You can also say that they are connected. Further, the connection relationship of the PWBA nodes in FIG. 3 follows the connection relationship of the PWBA nodes in FIG. 2, and the connection relationship is also determined by a failure inspection program or a failure inspection level.

  The PWBA (A) node 302, the PWBA (B) node 304, the PWBA (C) node 306, the PWBA (D) node 308, and the PWBA node (E) 310 are nodes indicating PWBA.

  Further, the FailCode (1) node 312 and the FailCode (2) node 314 are nodes indicating a fail code (Fail Code).

  Furthermore, the evidence information (L1) node 316, the evidence information (L2) node 318, and the evidence information (L3) node 320 are nodes indicating an evidence information node that represents the result of the failure state based on the inspection program or log.

  Then, the evidence information cooperation (L1) node 322 and the evidence information cooperation (L2) node 324 virtually link each information based on the evidence information node, analogize unknown information, and perform simulation. This is a node for performing fault diagnosis.

  Note that the PWBA (A) node 302, the PWBA (B) node 304, the PWBA (C) node 306, the PWBA (D) node 308, and the PWBA node (E) 310 physically use a plurality of raw resources based on the Bayesian network. It is an entity node that models a real entity. Further, the evidence information (L1) node 316, the evidence information (L2) node 318, and the evidence information (L3) node 320 are acquired from the result of determining the failure based on the history information output from the entity node, In contrast, the evidence information resource having no entity is modeled as a virtual node. Further, the evidence information linkage (L1) node 322 and the evidence information linkage (L2) node 324 mutually link the evidence information (L1) node 316, the evidence information (L2) node 318, and the evidence information (L3) node 320. However, it is modeled and generated as a cooperative node.

As described above, in the first embodiment, it is possible to accurately identify a failure location as a failure target from information such as evidence information and FailCode by generating a cooperation node.
(Second Embodiment)
Next, a second embodiment of the present invention will be described. In the second embodiment, the description of the same configuration as the first embodiment (particularly FIG. 1) is given the same reference numeral, and the description of the configuration is omitted.

  FIG. 4 is a diagram illustrating an example of an inspection program for inspecting a part of a control system configured by connecting a plurality of PWBAs as evidence information of a Bayesian network according to the second embodiment (diagram 400). ).

  In the chart 400, the inspection level is divided into three stages (inspection level (1), inspection level (2), and inspection level (3)), the meaning of each inspection level, the mode in which the inspection is performed, an example of inspection items, etc. Is shown.

  More specifically, at the inspection level (1), only the basic function is inspected so as not to disturb the activation operation. The inspection mode is performed in the startup mode and is performed at the time of startup every time. The inspection items are a mother board (Motor Board, hereinafter referred to as MB), a central processing unit (Central Processing Unit, hereinafter referred to as CPU), an application specific integrated circuit (Application Specific Integrated Circuit, hereinafter). ASIC), MB peripheral communication portion, memory module (also called memory module or expansion memory), network adapter (also called NIC (Network Interface Card) or LAN board or LAN card), exhaust heat blower And BIOS (abbreviation of Basic Input / Output System, a group of programs for controlling peripheral devices connected to the computer (Hereinafter, referred to as BIOS), only the operation state of whether or not each of the states operates is inspected.

  In the inspection level (2), an independent device (a computer or a peripheral device) that can be separated is inspected, and the inspection level (2) may be performed specifically for a specific part. The inspection mode is performed in the normal operation mode, and is performed when a machine is installed or an option is added. Inspection items include MB, CPU, ASIC, MB peripheral communication part, memory module, network adapter, function of image input device, and operation status of each function of facsimile in more detail than level (1). inspect.

  Further, at the inspection level (3), in addition to the inspection level (2), the entire inspection including the portion where the functions such as the mounted ASIC and the memory module are integrated is performed. The inspection mode is performed in the diagnosis-only mode, and is performed when a failure occurs. Inspection items are detailed operation states of all functions of MB, CPU, ASIC, MB peripheral communication part, memory module, network adapter, hard disk drive function, and image input device function from inspection level (2). Inspect deeply and in detail.

  Note that, for example, when viewing a physical place, it is called an area, and inspecting the same area, an inspection level (1), ( It is good also as 2) and (3). Further, for example, the inspection levels (1), (2), and (3) may be set in a format in which Log and FailCode for inspecting a certain area are prepared and a result of local inspection (local inspection Log) is output. . Furthermore, the same can be said not only when inspecting a physical location (hardware) but also when inspecting operating software (area for inspecting software) depending on the contents of the Log.

  FIG. 5 shows a PWBA connection configuration 500 to which the inspection program shown in FIG. 4 is applied, and each inspection level (inspection level (1), inspection level (2), inspection level when inspecting MB peripheral communication inspection items). (3)) shows an example of the range in which the inspection is actually applied.

  The PWBA connection configuration 500 includes PWBA (A) 502, PWBA (B) 504, PWBA (C) 506, PWBA (D) 508, PWBA (P) 510, and PWBA (Q) 512. As in FIG. 2, these plural PWBAs are physical entities and are called original resources.

  PWBA (A) 502 is connected to PWBA (B) 504, PWBA (B) 504 is connected to PWBA (C) 506, PWBA (P) 510, and PWBA (Q) 512, and PWBA (C) 506 is connected to PWBA (C) 506. (D) connected to 508.

  In addition, in each inspection level when the MB peripheral communication inspection item of the PWBA connection configuration 500 is inspected, the range to which the inspection is actually applied is the inspection range 530 of the inspection level (1), the inspection level as shown in the figure. The inspection range 540 of (2) and the inspection range 550 of the inspection level (3).

  In the case of FIG. 5, items to be inspected are communication inspections performed by two PWBAs, PWBA (B) 504 and PWBA (C) 506, via a 10-bit interface according to the inspection level. In the inspection level (1) range 530, PWBA (B) 504, PWBA (C) 506, and PWBA (D) 508, the communication between the three PWBAs and PWBAs, the main 2 bits are inspected. To do. In the range 540 of the inspection level (2), among the 10 bits of communication between the three PWBAs PWBA (A) 502, PWBA (B) 504, and PWBA (C) 506, 4 bits of communication and control signals are included. Conduct an inspection. In inspection level (3) range 550, PWBA (A) 502, PWBA (B) 504, PWBA (C) 506, PWBA (P) 510, PWBA (Q) 512, communication 10 bits between PWBA and PWBA Medium, 8-bit inspection of communication, control signal, and data line is performed.

  Note that there is a difference in the actual inspection area for each inspection level (see inspection level (1), inspection level (2), inspection level (3) in FIG. 4) for one inspection item. This is because there is a difference in the inspection method for each mode to be performed and the degree of detail (number of communication bits, number of elements on PWBA, function, etc.).

  FIG. 6 shows a failure diagnosis model 600 of the Bayesian network when the failure location is inferred by the Bayesian network based on FIG. 5, and the last two digits of each symbol correspond to each PWBA of FIG.

  The failure diagnosis model 600 includes a PWBA (A) node 602, a PWBA (B) node 604, a PWBA (C) node 606, a PWBA (D) node 608, a PWBA (P) node 610, a PWBA (Q) node 612, evidence information. A level (1) node 614, an evidence information level (2) node 616, an evidence information level (3) node 618, and an evidence information cooperation node 620 are configured.

  PWBA (P) node 610 is connected by an arc toward PWBA (B) node 604 and evidence information level (3) node 618, and PWBA (Q) node 612 is also connected to PWBA (B) node 604 and evidence information level (3). Connected by arc toward node 618. The PWBA (D) node 608 is connected to the PWBA (C) node 606 and the evidence information level (1) node 614 by an arc, and the PWBA (C) node 606 is connected to the PWBA (B) 604 and the evidence information level (1). ) Connected by arc towards node 614, evidence information level (2) node 616, and evidence information level (3) node 618. In addition, PWBA (B) node 604 is connected by an arc toward PWBA (A) 602, evidence information level (1) node 614, evidence information level (2) node 616, and evidence information level (3) node 618. Yes. The PWBA (A) node 602 is connected to the evidence information level (2) node 616 and the evidence information level (3) node 618 by an arc, and the evidence information level (1) node 614 and the evidence information level (2). The node 616 and the evidence information level (3) node 618 are connected to the evidence information cooperation node 620 by an arc.

  The operation of the second embodiment will be described below.

The six PWBAs (PWBA (A), PWBA (B), PWBA (C), PWBA (D), PWBA (P), PWBA (Q)) follow the actual connection configuration (see FIG. 5) and Contact Ri is expressed by attaching link causality as evidence information from the test result, it is linked against PWBA actually check is performed.

  It should be noted that, for example, the level of data (or quality) as the failure information, such as the level of inspection to be inspected when the device enters the power saving mode and returns, and to what inspection level information can be obtained. ), The evidence information level (1) node 614, the evidence information level (2) node 616, and the evidence information level (modeled based on the inspection levels (1), (2), and (3)). 3) Each node 618 etc. expresses. How much evidence information such as the evidence information level (1) node 614, evidence information level (2) node 616, and evidence information level (3) node 618 has been prepared, and how to deal with the evidence information It is necessary to consider that.

  Also, for example, one ASIC is simply inspected, but when 10 or more ASICs are mounted on a PWBA such as one motherboard, usually one or two are selected and inspected, Sometimes the entire ASIC is inspected. Alternatively, in the case of simple fault diagnosis, a signal is directly exchanged from the main PWBA only for fault diagnosis of an IC (Integrated Circuit) on a certain substrate. However, when performing failure diagnosis related to the overall operation, there are cases where cooperation with other PWBAs is involved, so the time and inspection range also vary depending on the level of failure diagnosis.

Therefore, it expresses the relationship between the evidence information that should be included in the diagnosis model, and the flexibility of the diagnosis system, for example, saving time and man-hours for failure, reducing the burden on customers, and ease of failure diagnosis, etc. Can improve diagnostic accuracy.
(Third embodiment)
Next, a third embodiment of the present invention will be described. In the third embodiment, the description of the same configuration as the first embodiment and the second embodiment (particularly FIG. 1) is given the same reference numerals, and the configuration of the configuration is the same. Description is omitted.

  Hereinafter, the operation of the third embodiment will be described based on the simulation results when a plurality of PWBAs 700 are specifically constructed.

  As shown in FIG. 7, the plurality of PWBAs 700 include PWBA (A) 710, PWBA (B) 720, PWBA (C) 730, PWBA (D) 740, PWBA (P) 750, and PWBA (Q) 760. It is configured.

  PWBA (A) 710 is connected to PWBA (B) 720, PWBA (D) 740 is connected to PWBA (C) 730, PWBA (C) 730 is connected to PWBA (B) 720, and PWBA (P) 750 And PWBA (Q) 760 is connected to PWBA (B) 720.

  In addition, in each inspection level when the inspection items of a plurality of PWBAs 700 are inspected, the range in which the inspection is actually applied is the inspection range 770 of the inspection level (1) and the inspection level (2) as illustrated. The inspection range is 780.

  Therefore, the inspection of the state of PWBA (B) 720 is performed in the form of inspection level (1) and inspection level (2). These inspection level (1) and inspection level (2) are substantially the same as those in FIG. It depends on the inspection mode in the description.

  Specifically, at inspection level (1), PWBA (A) 710, PWBA (P) 750, and PWBA (B) 720 for inspection of PWBA (B) 720, as indicated by inspection range 770 at inspection level (1), and Since the operation of PWBA (Q) 760 is required, the state of PWBA (A) 710, PWBA (B) 720, PWBA (P) 750, and PWBA (Q) 760 is substantially included in the inspection result. On the other hand, the inspection level (2) requires the operation of the PWBA (C) 730 for the inspection of the PWBA (B) 720 as indicated by the inspection range 780 of the inspection level (2) due to the same situation. Therefore, the state of PWBA (B) 720 and PWBA (C) 730 is substantially included in the inspection result.

  FIG. 8 is a Bayesian network 800 based on FIG. 7, and the last two digits of each symbol correspond to each PWBA of FIG. Bayesian network 800 includes PWBA (A) node 810, PWBA (B) node 820, PWBA (C) node 830, PWBA (D) node 840, PWBA (P) node 850, PWBA (Q) node 860, evidence information level. (1) Node 870, evidence information level (2) node 880, and evidence information cooperation node 890 are comprised.

  PWBA (P) node 850 is connected by an arc toward PWBA (B) node 820 and evidence information level (1) 870, and PWBA (Q) node 860 is connected to PWBA (B) node 820 and evidence information level (1) 870. Connected by arc towards. Also, the PWBA (D) node 840 is connected by arc toward the PWBA (C) node 830, and the PWBA (C) node 830 is connected by arc toward the PWBA (B) node 820 and the evidence information level (2) node 880. It is connected. Further, PWBA (B) is connected by arc towards PWBA (A) node 810 and evidence information level (2) node 880, and PWBA (A) is connected by arc towards evidence information level (1) node 880. The evidence information level (1) node 870 and the evidence information level (2) node 880 are connected by an arc toward the evidence information link 890.

  The operation of the third embodiment will be described below.

  Here, each inspection level (inspection level (1) and inspection level (2)) is expressed as an evidence information level by an evidence information level (1) node 870 and an evidence information level (2) node 880.

  Also, two evidence information level nodes (evidence information level (1) node 870 and evidence information level (2) node 880) treated as evidence information are used as parent nodes, and evidence information cooperation nodes that are child nodes of the parent node 890.

  Note that the inspection method is different at the inspection levels (1), (2), and (3) related to the above-described failure diagnosis. However, although inspection levels (1), (2), and (3) are the inspections shown in FIG. 4, they have the same common terms by changing the combination of hardware such as PWBA to be inspected for failure. Different inspections may be performed. For example, PWBA (B), PWBA (C), and PWBA (D) are inspected at inspection level (1), and PWBA (A), PWBA (B), and PWBA (C) are inspected at inspection level (2). Then, PWBA (B), PWBA (C), PWBA (D), and PWBA (Q) are inspected at the inspection level (3).

9 to 12 show the results of using the Bayesian network 800 of FIG. 8 with Bayesian network design software.
(First result)
First, a result of performing a failure inspection of a plurality of PWBAs using only initial information is shown as a first result 900 in FIG. When it is desired to diagnose a failure in the configuration of a plurality of PWBAs 700 shown in FIG. 7, performing two levels of inspection (inspection level (1) and inspection level (2)) requires time and manpower. Only inspection at inspection level (1) is performed.

At this time, the probability of NG (NO GOOD), which is the failure probability of the (initial) PWBA (B) node 820, is 1. 20 %.

Next, the state of the evidence information level (1) node 870 is determined by reflecting the result of the inspection level (1) (hereinafter, the determination of the state of the node is referred to as activation). Activation also means that detailed information results are specifically applied to the node. Specifically, to make the probability of NG 100% is to activate. The result of activating the evidence information level (1) node 870 is shown in FIG.
(Second result)
FIG. 10 shows a second result 1000 that actually simulates the result of activating the evidence information level (1) node 870 (see double frame A in FIG. 10) using the Bayesian network 800 of FIG. .

  Referring to the result, the NG probability of the PWBA (B) node 820 changes to 17.2%.

Next, similarly to the above, when the state of the evidence information cooperation node 890 is activated and the probability of NG is 100%, a result as shown in FIG. 11 is shown.
(Third result)
FIG. 11 uses the Bayesian network 800 of FIG. 8 and uses the evidence information level (1) node 870 (see the double frame A in FIGS. 10 and 11) and the evidence information cooperation node 890 (see the double frame B in FIG. 11). A third result 1100 obtained by actually simulating the result obtained by activating X is shown.

  When attention is paid to the evidence information cooperation node 890 as shown in FIG. 11, if the evidence information results are not complete, the evidence information cooperation node 890 is activated to make the failure tendency clear.

  In FIG. 10, by activating the evidence information level (1) node 870, if the result of the inspection level (1) is determined to be NG as failure, the evidence information cooperation node 890 is set to NG as shown in FIG. When activated, the NG probability of the PWBA (B) node 820 increases to 34.1%. However, when the evidence information cooperation node 890 is activated to be normal OK, the NG probability of the PWBA (B) node 820 falls to 8.91%, although not shown. This indicates a failure tendency of the PWBA (B) node 820, which is a common item between the inspection level (1) and the inspection level (2).

  That is, when the result becomes NG at the inspection level (1), the possibility that the PWBA (B) node 820 becomes NG changes from 17.2% to 34.1%, so that it is simply By using the composition of the evidence information, in this case the relationship between the two evidence information of the inspection level (1) and the inspection level (2), as the result of the inspection level (1) is not only 17.2% Can be analogized.

  On the other hand, for example, if attention is paid to the state of PWBA (P), the probability of 13.2% NG is obtained based on the result of the inspection level (1) (see FIG. 10). By doing so, the probability of NG is corrected downward to 12%. However, although not shown, by setting the evidence information cooperation node 890 to be OK, the probability of NG becomes 15.2% and is corrected upward.

Therefore, by operating the evidence information cooperation node 890 (in this case, by operating only NG or OK), it is possible to improve the accuracy of the probability of detecting the failure of the device. More specifically, limited evidence information (for example, evidence information level (1) node 870 and evidence information in FIG. 10, FIG. 11, and FIG. 12) indicating which device node has failed at the stage when the causal relationship is established. By operating the evidence information cooperation node 890 from all information in the level (2) node 880), the accuracy of the probability of detecting a device failure by analogizing unclear evidence information is improved. It can also be said that it is possible.
(Fourth result)
FIG. 12 uses the Bayesian network 800 of FIG. 8 and uses the evidence information level (1) node 870 (see the double frame A in FIGS. 10, 11 and 12) and the evidence information cooperation node 890 (FIGS. 10 and 11). And a fourth result 1200 obtained by actually simulating the result of activating the evidence information level (2) node 880 (see the double frame C in FIG. 12).

  Table 1 shows the probability of NG when the evidence information level (1) node 870, the evidence information linkage node 890, and the evidence information level (2) node 890 are set to NG in order from the inactive state (inactive state). Indicates.

従って、証拠情報間（証拠情報レベル（１）ノード８７０及び証拠情報レベル（２）ノード８８０の間）の関係を明示した証拠情報連携ノード８９０を用意し、証拠情報間に階層性を持たせたベイジアンネットワークを構成し、証拠情報連携ノード８９０を操作することにより、その場ですぐに知ることができる証拠情報のみから故障の傾向を知ることができるため、時間や工数を節約しながら、故障確率の精度を上げることができる。

Therefore, the evidence information cooperation node 890 that clearly shows the relationship between the evidence information (between the evidence information level (1) node 870 and the evidence information level (2) node 880) is prepared, and the evidence information has a hierarchy. By configuring the Bayesian network and operating the evidence information linkage node 890, the failure tendency can be found only from the evidence information that can be immediately known on the spot, so the failure probability can be saved while saving time and man-hours. Can improve the accuracy.

  Furthermore, even if a certain part of a plurality of pieces of evidence information is missing, the missing part has default information that is preset information, and based on the default information, specifically, Is assumed in a specific direction as to what kind of situation it is, and it is possible to predict which direction the situation to be diagnosed is facing.

  Further, in such a situation, when a failure occurs immediately after the device is activated, the failure state of the device can be grasped without taking time for detailed diagnosis. Or, after a long time has passed since startup, it can be a wide variety of effective means, such as when a device failure occurs and you want to grasp the tendency of failure possibility without the need to restart again .

  5 and 6, the communication state between a certain PWBA (between PWBA (B) 550 and PWBA (C) 560), and in the third embodiment, a certain PWBA (PWBA (B) 720). Although an example of diagnosing the state of the present invention has been described, the present invention is not limited to them and can be used in many situations. For example, diagnostics in units of ICs such as ASICs and CPUs installed in PWBA, diagnostics that apply diagnostic targets to software applications, hard disk drives (Hard Disk Drives) and DIMM modules (Dual Inline Memory Modules) specialized on motherboards The same concept can be applied to diagnosis, diagnosis related to version information, and a system constituted by a plurality of PWBA groups via a network.

  In the third embodiment, the evidence information nodes (evidence information level (1) node 870 and evidence information level (2) node 880) which are the parent nodes of the evidence information cooperation node 890 are based on the local inspection results. However, in practice, all evidence information that may cause a failure, such as environmental information such as temperature and humidity, failure code information prepared in the device, and the like is applied.

  Furthermore, although each evidence information node has only two state transitions of OK and NG, examples of states that can actually be taken can be classified into some conditions.

  Further, in the third embodiment, for the sake of simple explanation, it is expressed and described in the form of one evidence information cooperation node 890. However, depending on the content of the evidence information expressed in the failure diagnosis model, a plurality of pieces of evidence information are provided. There can also be cooperative nodes.

  Furthermore, including the case where there are a plurality of evidence information cooperation nodes, one evidence information cooperation node can be applied as a parent node, and the remaining evidence information cooperation nodes can be applied as child nodes. In this case, the evidence information cooperation node can also use the evidence information node as a parent node. A plurality of evidence information cooperation nodes can be used as parent nodes, and then one evidence information cooperation node can be used as a child node.

It is the schematic which shows the whole structure about the network for applying the failure diagnosis which concerns on 1st Embodiment, and the apparatus connected to it. FIG. 2 is a simple block diagram of a configuration of a PWBA that controls the device of FIG. 1. It is a model figure at the time of inferring a failure location by the Bayesian network which is a causal network based on FIG. 10 is a chart showing an example of an inspection program for inspecting a part of a control system configured by connecting a plurality of PWBAs for applying failure diagnosis according to the second embodiment. FIG. 5 is a PWBA connection configuration diagram to which the inspection program shown in FIG. 4 is applied. FIG. 6 is a fault diagnosis model diagram for inferring a fault location by a Bayesian network based on FIG. 5. It is a block diagram of the simulation based on 3rd Embodiment and based on several PWBA. It is a Bayesian network block diagram based on FIG. It is explanatory drawing which shows the 1st result based on FIG. It is explanatory drawing which shows the 2nd result based on FIG. It is explanatory drawing which shows the 3rd result based on FIG. It is explanatory drawing which shows the 4th result based on FIG.

Explanation of symbols

210, 540 PWBA (A) (raw resources)
220, 550 PWBA (B) (raw resources)
230, 560 PWBA (C) (raw resources)
240, 570 PWBA (D) (raw resources)
250 PWBA (E) (raw resources)
302, 602, 710, 810 PWBA (A) node (substance node)
304, 604, 720, 820 PWBA (B) node (substance node)
306, 606, 730, 830 PWBA (C) node (substance node)
308, 608, 740, 840 PWBA (D) node (substance node)
310 PWBA (E) node (substance node)
316 Evidence information (L1) node (virtual node)
318 Evidence information (L2) node (virtual node)
320 Evidence information (L3) node (virtual node)
322 Evidence Information Cooperation (L1) Node (Cooperation Node)
324 Evidence Information Cooperation (L2) Node (Cooperation Node)
580 PWBA (P) (raw resources)
590 PWBA (Q) (raw resources)
610, 750, 850 PWBA (P) node (substance node)
612, 760, 860 PWBA (Q) node (substance node)
614, 870 Evidence information level (1) Node (virtual node)
616, 880 Evidence information level (2) Node (virtual node)
618 Evidence information level (3) Node (virtual node)
620, 890 Evidence Information Cooperation Node (Cooperation Node)

Claims (4)

Node generation means for generating a node by modeling at least one of hardware or software that is a plurality of original resources existing as functionally necessary resources;
A failure determination unit that associates and connects the plurality of nodes to be subjected to failure diagnosis, and determines failure based on history information output from a system represented by the plurality of nodes;
Virtual node generation means for generating a virtual node obtained by modeling from the determination result by the failure determination means, unlike at least one of the hardware and software that is the original resource, modeling an evidence information resource that does not exist.
A plurality of virtual nodes that cooperate with each other, model them, and generate a cooperation node;
Analogy means for analogizing evidence information of the virtual node by the cooperation node;
A plurality of virtual nodes whose evidence information has been clarified by analogy by the analogy means, and a specifying means for identifying a faulty node by changing each of the determination balances of at least pass / fail which both the cooperation nodes show ,
Having a fault diagnosis system.   The failure diagnosis system according to claim 1 is executed by a failure diagnosis program based on a Bayesian network, and the plurality of virtual nodes and the cooperation node serve as a unit for specifying a failure when the failure diagnosis program is executed. And fault diagnosis system. Computer
Node generation means for generating a node by modeling at least one of hardware or software that is a plurality of original resources existing as functionally necessary resources;
A failure determination unit that associates and connects the plurality of nodes to be subjected to failure diagnosis, and determines failure based on history information output from a system represented by the plurality of nodes;
Virtual node generation means for generating a virtual node obtained by modeling from the determination result by the failure determination means, unlike at least one of the hardware and software that is the original resource, modeling an evidence information resource that does not exist.
It said plurality of virtual nodes mutually cooperate and modeled, and the coordination node generating means for generating a coordination node,
Analogy means for analogizing evidence information of the virtual node by the cooperation node;
It said plurality of virtual nodes evidence information by analogy by the analogy means revealed by changing respective determination balance of at least quality both indicate the coordination node, specifying means for specifying a node has failed ,
Fault diagnosis program to function as   The failure diagnosis program according to claim 3 is executed by a failure diagnosis program based on a Bayesian network, and the plurality of virtual nodes and the cooperation node serve as a unit for specifying a failure when the failure diagnosis program is executed. Fault diagnosis program.

Images