JP2008256981A - Fault diagnostic system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a fault diagnostic system capable of more accurately diagnosing the cause of fault by eliminating influence of incorrect information included in an instance database. <P>SOLUTION: The fault diagnostic system is equipped with: a market quality information database 40 for storing the obtainable number of fault phenomena, fault parts, fault states and disposed content related to cause of fault for every machine kind of the image forming apparatus 10; a diagnostic model updating part 34 acquiring the obtainable number of all the combinations of the fault part related to the cause of the fault with at least one of the fault phenomenon, the fault state and the disposed content related to the cause of the fault for every cause of the fault from the market quality information database 40; and a probability updating processing part 33 calculating conceivable probability among the respective nodes of a Baysian network showing the fault diagnostic models of components constituting the image forming apparatus 10 based on the number of combinations of every cause of the fault acquired by the diagnostic model updating part 34. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a failure diagnosis system.

  Conventionally, in office equipment such as copiers and printers, specialized service personnel have been dispatched and regularly maintained in order to maintain good quality. However, with the recent trend toward color and high functionality of office equipment, the state of failure has become more complex, and even a professional service person may not be able to identify the cause of the failure. Furthermore, since it is necessary to minimize the downtime of equipment on the customer side, there are many cases in which parts that are likely to be related to failure are replaced together. For this reason, there is a problem in that normally normal parts are also replaced together, resulting in an increase in service cost.

In response to this problem, we have applied for a fault diagnosis method (see Patent Document 1) for estimating a fault location using a Bayesian network. The Bayesian network is given in advance an initial probability based on a database in which failure occurrence frequencies in the market are accumulated for the failure cause node. Then, the number of occurrences of each failure cause is automatically acquired periodically from the database and the initial probability of the Bayesian network model is updated, so that diagnosis based on the latest failure occurrence frequency in the market becomes possible. ing.
Japanese Patent Laying-Open No. 2005-309078

  However, the information stored in the database consists of information entered by a person who has performed maintenance such as a service person for each model after troubleshooting. For this reason, some treatments were performed, but the original cause was unknown because multiple sites were exchanged at once, “appropriate treatment could not be performed”, “code information of the site where treatment was performed was predecessor Despite being different from that of the machine, the code information of the previous machine is used for the new machine report without noticing it. Is also included. Therefore, when the initial probability is calculated using the information in the database as it is, there is a problem that the diagnosis system is deteriorated because it includes an error with respect to the actual situation.

  The present invention has been made in view of the above-described facts, and an object of the present invention is to provide a failure diagnosis system capable of diagnosing a cause of failure with higher accuracy, excluding the influence of erroneous information included in a case database.

  In order to solve the above-described object, the failure diagnosis system according to claim 1 is configured to calculate, for each model of the image forming apparatus, the number of occurrences of each of the failure phenomenon, the failure site, the failure state, and the treatment content related to each failure cause. For each failure cause, the storage means stores the number of occurrences of all combinations of the failure part related to the failure cause and at least one of the failure phenomenon, the failure state, and the treatment content related to the failure cause. And an occurrence probability between each node of the Bayesian network representing a failure diagnosis model of a part constituting the image forming apparatus, based on the acquisition unit acquired by the acquisition unit and the number of occurrences for each failure cause acquired by the acquisition unit. Occurrence probability calculating means.

  The failure diagnosis system according to claim 2 is the failure diagnosis system according to claim 1, wherein the occurrence probability calculation means includes the failure phenomenon corresponding to the failure cause in the number of occurrences of all combinations acquired by the acquisition means, A multiplying unit that multiplies a weighting factor indicating the degree of relevance of each of the failure part, the failure state, and the treatment content; and a sum total calculating unit that calculates the total number of occurrences of all combinations of the results multiplied by the multiplying unit. It is characterized by.

  The failure diagnosis system according to claim 3 is the failure diagnosis system according to claim 2, wherein the acquisition unit further includes all the failure portions of other image forming apparatuses used before the image forming apparatus to be diagnosed. The multiplication means obtains a weighting factor indicating the degree of relevance of the failure part of the other image forming apparatus with respect to the cause of the failure to the occurrence number of all combinations acquired by the acquisition means. Furthermore, it is characterized by multiplication.

  The failure diagnosis system according to claim 4 is the failure diagnosis system according to any one of claims 1 to 3, wherein the failure cause of the component is determined based on the occurrence probability calculated by the occurrence probability calculation means. It further comprises an inference means for inferring.

  As described above, according to the first aspect of the present invention, since the number of occurrences is calculated based on not only the failure part but also information on the failure phenomenon, the failure state, and the action content, the occurrence probability is obtained. Can be obtained.

  According to the invention described in claim 2, since the number of occurrences is calculated based on the weighting factor set for each item of failure phenomenon, failure site, failure state, and action content, the influence of information with low reliability is excluded. The effect that the diagnosis becomes possible is obtained.

  According to the invention described in claim 3, since the information of the predecessor machine is held in addition to the model to be diagnosed, it is possible to perform diagnosis without the influence of erroneous information such as the information of the predecessor machine being used incorrectly. An effect is obtained.

  According to the fourth aspect of the invention, there is an effect that the cause of the failure can be inferred with high accuracy for each image forming apparatus.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a basic configuration of an image forming apparatus 10 that constitutes a failure diagnosis system according to the present invention. The image forming apparatus 10 includes an image reading unit 11 that reads a document image, a print engine unit 12 that forms and outputs a read image or a print-instructed image, paper passage time, drive current, internal temperature and humidity, and the like. A sensor unit 13 for obtaining internal state information of the device, a diagnostic information input unit 14 for a user to input information necessary for failure diagnosis, and a failure diagnosis unit 15 for performing device failure diagnosis based on each acquired information And a communication unit 16 that obtains the latest diagnostic model from a management device that is a server computer via a network.

  FIG. 2 is a functional block diagram of the failure diagnosis unit 15 constituting the image forming apparatus 10 described above. The failure diagnosis unit 15 includes a component state information acquisition unit 21 that acquires component information indicating the operation state of each component acquired by the sensor unit 13 as observation data information, and a monitoring result of the usage status of the image forming apparatus 10 as history information. The user inputs the history information acquisition unit 22 to be acquired as an environmental information acquisition unit 23 to acquire the internal environmental conditions directly or via the sensor unit 13, and the diagnostic information input unit 14 through the diagnostic information input unit 14. Information obtained from each of the acquisition units 21 to 25, a failure information acquisition unit 24 that acquires failure information obtained by the above, an additional operation information acquisition unit 25 that acquires failure information in a different operating condition depending on a user operation, and A failure probability inference unit 26 that calculates the probability of failure based on the diagnosis result notification unit 27 that notifies the user of the diagnosis result is provided.

  Further, the failure probability inference unit 26 is calculated by an inference engine 28 that calculates the probability that each failure cause candidate that is considered to cause a failure is the main cause of the failure that has occurred, based on the acquired information, and the inference engine 28. And a failure candidate extraction unit 29 that narrows down failure cause candidates based on the probability.

  Here, a Bayesian network is used for the inference engine 28 that calculates the occurrence probability of the failure cause. A Bayesian network represents a problem area where the causal relationship is complex. Therefore, the causal relationship between multiple variables is sequentially connected and expressed as a network with a graph structure, and the dependency relationship between variables is expressed by a directed graph. .

  FIG. 3 shows a configuration example of a Bayesian network in the case of performing a fault diagnosis of an image defect system. As shown in the figure, this Bayesian network includes a failure cause node ND0 that represents a cause of causing an image defect, a component state node ND1 that represents state information of components (components) constituting the image forming apparatus, and the image forming apparatus. History information node ND2 representing history information, environment information node ND3 representing ambient environment information where the image forming apparatus is installed, observation state node ND4 representing image defect state information, and follow-up result information obtained by user operation A user operation node ND5 representing a defect type node ND6.

  Each node is connected so as to have a relationship of “cause” → “result”. For example, the relationship between the “failure cause node ND0” and the “observation state node ND4” is a relationship in which “observation state (light density, streaks / bands, etc.)” appears based on the “failure cause”. On the other hand, the relationship between the “history information node ND2” and the “failure cause node ND0” is “failure cause (part deterioration, etc.) based on“ state based on history information (number of copies, long operation years, etc.) ”. "Is generated. Details of the failure diagnosis method using the Bayesian network are described in Patent Document 1.

  FIG. 4 shows a specific example of the failure diagnosis system, and represents a Bayesian network when a “line / band” occurs in the failure diagnosis configuration example due to the image defect shown in FIG. As shown in the figure, the nodes are connected so as to have a relationship of “cause” → “result”. For example, the relationship between “drum scratches” and “line width information” is such that “line width information” such as the occurrence of thin lines based on “drum scratches” appears. On the other hand, the relationship between “feed number history information” and “fuser” is highly likely to cause lines and bands due to deterioration of “fuser” based on the state based on “feed number history information” (number of feeds or more). The relationship that

  The initial value of the probability data of each node is determined based on past data. After that, the probability of each node is updated regularly (for example, every month) based on statistical data of market troubles such as part replacement frequency and defect occurrence frequency. Further, gray nodes indicating features of image defects such as “line width information”, “periodic information”, and “occurrence location information” are “line-based” based on the feature amounts obtained by the failure information acquisition unit 24 of FIG. If it is “width information”, a failure state such as “thin line is generated” is determined.

  FIG. 5 shows a configuration example of a failure diagnosis system according to the present invention. In the failure diagnosis system, the market quality information input device 50, the market quality information database 40, the management device 30, and the image forming device 10 are connected via a network. The management device 30 acquires from the market quality information database 40 data necessary to update the occurrence probability of the failure cause node constituting the diagnosis model and the diagnosis model storage unit 31 that stores the diagnosis model used for failure diagnosis. And an update information management unit 32 that manages attribute information necessary for updating, and a probability update processing unit 33 that calculates and updates the occurrence probability of each failure cause node based on the information held by the update information management unit 32 The diagnostic model is updated by reflecting the probability calculated by the probability update processing unit 33 in the diagnostic model, and is stored in the diagnostic model storage unit 31, and is updated from the market quality information database 40. And a communication unit 35 that receives information and communicates with the image forming apparatus 10 to transmit the updated diagnostic model.

  The market quality information database 40 stores, as code information, detailed items relating to failure phenomena, failure sites, failure states, and treatment details of troubles occurring in the market in association with the number of occurrences. Then, the diagnostic model update unit 34 can know the number of occurrences of the corresponding content by designating one or a plurality of codes. The market quality information input device 50 inputs the above various codes after a serviceman handles a trouble in the market and registers them in the market quality information database 40.

  6 and 7 show the relationship between the code information of each item registered in the market quality information database 40 and the weighting factor for the failure cause to be updated. The weight coefficient information is held in the update information management unit 32 of the management device 30 and is used by referring to the update information management unit 32 when the probability update processing unit 33 executes the probability update process. The weight coefficient information is provided for each node that does not have a parent node as a failure cause node of the diagnostic model. In the example of FIG. 4, weight coefficient information is set for each failure cause node surrounded by a double line. FIG. 6 shows the weighting factor information when the failure cause node is “drum flaw” with respect to the Bayesian network when a “line / band” of a certain model to be diagnosed occurs. The weight coefficient information is set for each item of the failure phenomenon, the failure site, the failure state, and the treatment content shown in FIGS.

  First, since the failure phenomenon is a “line / band” diagnostic model, code information and a weighting coefficient for the item “line / band” are set. Furthermore, in the present embodiment, a tandem type copying machine is assumed as an image forming apparatus to be diagnosed for failure. In this case, “drum scratches” usually occur in a single color, and the frequency with which multiple colors occur simultaneously. Is low. For this reason, as shown in FIG. 6A, the weighting factor is set to 1.0 for each failure phenomenon of yellow (Y), magenta (M), cyan (C), and black (K), and color mixing is performed. 0.1 is set for this failure phenomenon.

  Next, for the failed part, since the target part of “drum scratch” is “Drum Cartridge” of each color, code information and a weighting factor of 1.0 are set for each. Normally, the code information of the part of the model to be diagnosed may be set here. However, in the case of a part that is frequently maintained, such as “Drum Cartridge”, the service person can trouble with the market quality information input device 50. When inputting information, there is a case where the code information of parts of other models (predecessor) is remembered and input as it is. In this case, when the diagnostic model update unit 34 acquires information on the model to be diagnosed from the market quality information database 40, for example, for the part “Drum Cartridge”, the codes of other models other than the number of code information of the model to be diagnosed The number of cases by information is also included. As a result, if attention is paid only to the number of cases based on the code information of the model to be diagnosed, a deviation from the actual number of occurrences occurs, and the diagnosis accuracy decreases. Therefore, as shown in FIG. 6B, the weight coefficient is set by adding not only the part code information of the model to be diagnosed but also the part code of another model to “Drum Cartridge”.

  Next, the failure state is shown in FIG. 6C with respect to the failure state related to the “drum scratch” that generates the “line / band” among the various failure state information registered in advance. The weight coefficient is set as follows. The gray part in FIG. 6 represents items that are not related to “drum scratches” among the items registered as the failure state information, and the weighting factor is described as 0 for explanation. Actually, it is not held in the update information management unit 32 of the management device 30. Thus, the weighting coefficient is set only for the related items, and the other items are not held in the update information management unit 32 of the management device 30. This is because, when a service person inputs trouble information using the market quality information input device 50, there is a case where an erroneous input is made or a corresponding part that cannot be properly dealt with is registered for the time being. This is to eliminate as much as possible items that can be misinformation.

  Next, with respect to the treatment content, among the various treatment content information registered in advance, the treatment content that can be taken when the “drum scratch” occurs, as shown in FIG. Set. Here, the setting method of the gray portion and the weighting coefficient is the same as in the case of the above-described failure state.

  FIG. 7 shows weighting factor information when the failure cause node is “dirt of drum” with respect to a Bayesian network when a “line / band” of a model to be diagnosed occurs.

  The method of setting the failure part and the weighting coefficient for the failure part shown in FIGS. 5A and 5B is the same as the above “drum flaw”, but “drum fouling” compared to “drum flaw”. Since the frequency of occurrence of a plurality of colors at the same time is higher, the weight coefficient of the failure phenomenon “line / band (mixed color)” in FIG. 7A is higher than that in FIG. 6A.

  In FIG. 6C, the failure condition is 1.0 for the “scratch” weight coefficient, but does not cause “dirt of the drum”. 32 does not hold the information. For other items, the weighting coefficient is set as shown in FIG. 7C in accordance with the degree of association with “drum dirt”. Similarly, for the treatment content, a weighting coefficient is set as shown in FIG.

  Next, a method of updating the occurrence probability of the failure cause node constituting the diagnostic model using the weight coefficient information described in FIGS. 6 and 7 will be described with reference to the flowchart shown in FIG.

  First, in step 100, the diagnostic model update unit 34 of the management device 30 selects a diagnostic model to be subjected to probability update processing from the diagnostic model group stored in the diagnostic model storage unit 31.

  In step 110, the diagnostic model update unit 34 refers to the weighting factor information held in the update information management unit 32, and refers to the failure phenomenon, failure part, failure corresponding to the failure cause node constituting the selected diagnosis model. The code information of the status and treatment content is acquired, and the number of occurrences for each set of information is acquired from the market quality information database 40.

  When the cause of failure in FIG. 6 is “drum scratch”, the following processing is performed. First, the diagnostic model update unit 34 corresponds to a set of [failure phenomenon, failure site, failure state, treatment content] [wire / band (Y), Drum Cartridge (Y) -target machine, wear, trouble replacement]. Code information [1001, 2001, 3001, 4001] is transmitted to the market quality information database 40 together with the probability update target model information, and the number of occurrences of the target model for this set is acquired. Next, code information [1001, 2001, 3001, 4004] corresponding to [line / band (Y), Drum Cartridge (Y) -target machine, wear, first aid] is transmitted to the market quality information database 40, and the target Get the number of occurrences for this set of models. In this way, the number of occurrences for all combinations of items reflected in the weight coefficient information is acquired.

  In step 120, the probability update processing unit 33 multiplies the number of occurrences of each set acquired in step 110 by a corresponding weight coefficient. For example, in FIG. 6, the set of [failure phenomenon, failure site, failure state, treatment content] is [number of lines / bands (Y), Drum Cartridge (Y)-target machine, wear, trouble replacement] If there are 200, the corresponding weighting factors, 1.0, 1.0, 0.3, and 1.0 are sequentially multiplied. Therefore, the number of occurrences of the set [line / band (Y), Drum Cartridge (Y) -target machine, wear, trouble replacement] reflecting the weighting factor is 200 × 1.0 × 1.0 × 0.3 ×. 1.0 = 60 cases. Similarly, the number of occurrences reflecting the weighting factor is calculated for other groups.

  In step 130, the probability update processing unit 33 accumulates the multiplication results in step 120 and sets the sum as the number of occurrences for the failure cause node. The processing from step 110 to step 130 is expressed by equation (1).

Here, N _C is the number of occurrences of failure cause nodes to be calculated, α _i (i = 1 to m) is a weighting factor of the i-th failure phenomenon, and β _j (j = 1 to n) is the j-th failure. The weighting factor of the part, γ _k (k = 1 to o) is the weighting factor of the k-th failure state, θ _l (l = 1 to p) is the weighting factor of the first treatment content, and N _ijkl is the i-th failure. This represents the number of occurrences for a set of phenomenon, j-th failure site, k-th failure state, and l-th treatment content.

  In step 140, it is determined whether the probability update processing unit 33 has calculated the number of occurrences for all failure cause nodes constituting the diagnostic model. If the determination is affirmative, the process proceeds to step 150. If the determination is negative, the process returns to step 110 and the processes from step 110 to step 130 are repeated.

  In step 150, the probability update processing unit 33 calculates an occurrence probability for each failure cause node based on the calculated number of occurrences of all the failure cause nodes. The method for calculating the occurrence probability will be described using the diagnosis model shown in FIG. 4 as an example.

  As described above, the failure cause node having the weight coefficient information is represented by a double line in FIG. 4, and the number of occurrences for the node surrounded by the double line is calculated up to step 140. The occurrence probability is calculated between nodes having common child nodes.

  For example, since the child node for the “drum flaw” node is “drum unit” and the failure cause node having the same child node is the “drum dirt” node, the probability of occurrence between the two nodes is calculated. If the number of occurrences of “drum scratches” is 60 and the number of occurrences of “drum stains” is 40, the probability of “drum scratches” occurring at the “drum unit” node is 60%, and “drum stains” are occurring. The probability is 40%. Then, this result is reflected in the conditional probability table of the “drum unit” node which is a child node.

  FIG. 9 shows an example of a conditional probability table. The table shown in the drawing represents the relationship between the “drum unit” node as a child node and the “drum scratch” node and the “drum dirt” node as its parent nodes. The table element represents the probability that the failure cause of the child node is normal or abnormal when the failure cause of each parent node is normal or abnormal.

  Specifically, in the conditional probability table, when “drum scratch” and “drum dirt” of the parent node are normal (not the cause of failure), the child node is caused by any abnormality of the parent node. The probability that the “drum unit” is abnormal is 0%. That is, in this state, the child node “drum unit” is normal with a probability of 100% (second row and second column in the table of FIG. 9).

  Also, the conditional probability table shows that if the “drum scratch” of the parent node is abnormal and the “dirt dirt” is normal, the child node is caused by the “drum scratch” of the parent node. The probability that a certain “drum unit” is abnormal is 60% (the third row and the second column in the table of FIG. 9).

  Further, the probability update processing unit 33 similarly sets conditional probabilities for other failure cause nodes. For example, the number of occurrences of “heat roll scratches” and “offset” is used for “fuser”, and the number of occurrences of “board”, “drum unit”, and “fuser” is used for “image output system”. Calculate and set the probability using. Note that the number of occurrences of failure causes such as “drum unit” and “fuser” that are not double lines is calculated as the total number of occurrences of the parent node. In the present embodiment, the number of occurrences of “drum units” is 60 + 40 = 100.

  In addition, the initial probability of the failure cause node surrounded by a double line having no parent node is set to a value close to 100% by the probability update processing unit 33 as a normal state (for example, 99.9%). Such). By setting in this way, in the state where no failure has occurred, that is, in the state where evidence information (information based on the feature amount obtained by the failure information acquisition unit 24 in FIG. 2) is not given, The probability is almost 0%. When a failure occurs and evidence information is given, the probability of each failure cause is recalculated by the probability update processing unit 33 by probability propagation.

  In step 160, the diagnostic model updating unit 34 updates the conditional probability table of each failure cause node of the diagnostic model based on the conditional probability of each failure cause node calculated as described above, and the diagnostic model storage unit 31.

  In step 170, the diagnostic model update unit 34 determines whether or not the conditional probability tables of all diagnostic models stored in the diagnostic model storage unit 31 have been updated. If a negative determination is made, the process returns to step 100 and the processes from step 100 to step 160 are repeated. If an affirmative determination is made, the probability update process is terminated.

  As described above, by calculating the conditional probability of the failure cause node using the weight coefficient information, it is possible to eliminate erroneous information contained in the market quality information database 40 as much as possible and to obtain the failure cause occurrence probability with high accuracy. it can. In addition, for rare cases such as rare cases, if the number of erroneous information is included, the error in the probability of occurrence will increase, and it will greatly affect the diagnosis result.By applying the present invention, A more accurate occurrence probability can be obtained.

  Further, the failure diagnosis unit 15 provided in the image forming apparatus 10 diagnoses the cause of failure with higher accuracy by inferring the cause of failure of the component based on the probability of occurrence of the cause of failure thus obtained. Is possible.

  In addition, a case where erroneous information affects the diagnosis result will be described below. FIG. 10 shows an example in which the occurrence probability of the failure cause is reflected when the erroneous information is included in the Bayesian network of FIG. 4 and when the erroneous information is removed. In FIG. 10, in order to simplify the description, only the failure cause node (double line) and the failure cause intermediate node (single line) are common to FIG. 4, and two observation state nodes are provided.

  Here, for example, assume that 15 out of 1005 cases have occurred as a result of obtaining the number of occurrences without performing the processing from step 110 to step 130 in FIG. In this case, the initial probability of “platen scratch” is 15 / 1005≈1.5%.

  Similarly, as a result of calculating initial probabilities of other failure-causing nodes, it is assumed that the values in the upper stage of the numerical values shown in the second row are in the vicinity of each failure-causing node in FIG. Further, the probability described below is provided in the vicinity of the “observation state 1” node and the “observation state 2” node in FIG. 10, and the occurrence probability of each failure cause when the state of these two nodes is determined as evidence information. When is calculated, the following results are obtained in the descending order of probability (4th place is omitted).

1. Platen scratch: 30.3%
2. Board B failure: 28.7%
3. Board A failure: 22.9%
Next, assume that 10 cases have occurred as a result of obtaining the number of occurrences by performing the processing from step 110 to step 130 in FIG. In other words, the five differences are misinformation included in the gray part of FIG. 6 and the number of occurrences of other failure cause nodes does not change for the sake of simplicity. Become. At this time, the initial probability of the “platen scratch” is 10/1000 = 1%.

  Similarly, if the initial probabilities of other failure-causing nodes are recalculated with the denominator being 1000, the lower values shown in the second row in the vicinity of each failure-causing node in FIG. When the occurrence probability of each cause of failure when the evidence information is given to the “observation state 1” node and the “observation state 2” node using these values, the following results are obtained in descending order of probability (4 Omitted below).

1. Board B failure: 31.8%
2. Board A failure: 25.4%
3. Platen scratches: 22.8%
As can be seen from the above results, if erroneous information is included in setting the initial probability in the failure cause node, the result of diagnosis using the Bayesian network is different from the result that should be obtained and the diagnosis accuracy is significantly reduced. End up. In the above example, for the sake of simplicity, a comparison was made between a case where erroneous information was included in only one node and a case where erroneous information was not included. However, when erroneous information is also included in other nodes, the error further increases. . Therefore, as described above, when calculating the conditional probability of the failure cause node, more accurate failure diagnosis is possible by eliminating the erroneous information contained in the database as much as possible and obtaining the failure cause occurrence probability with high accuracy. Is possible.

  The present invention is not limited to the embodiment described above. For example, the present invention may be applied not to a Bayesian network but to an expert system, a table reference type diagnosis system, and other diagnosis systems using AI.

1 is a basic configuration diagram of an image forming and manufacturing apparatus constituting a failure diagnosis system according to the present invention. FIG. 2 is a functional block diagram of a failure diagnosis unit constituting the image forming apparatus of FIG. 1. It is a block diagram of a Bayesian network when performing a fault diagnosis of an image defect system. It is a block diagram of the Bayesian network at the time of line | wire / band generation | occurrence | production in the structural example of the failure diagnosis by an image defect. 1 is a configuration diagram of a failure diagnosis system according to the present invention. FIG. It is a figure which shows the example of a setting of the weighting coefficient with respect to the failure cause "drum flaw" of the diagnostic model at the time of line | wire / strip | belt generation | occurrence | production. It is a figure which shows the example of a setting of the weighting coefficient with respect to the cause of failure "drum dirt" of a diagnostic model at the time of line | wire / belt | band | zone generation | occurrence | production. It is a flowchart which shows the flow of the update process of the generation probability of the failure cause node which comprises a diagnostic model. It is a figure which shows the example of a conditional probability table. It is a figure which shows the example which reflected the generation | occurrence | production probability of the cause of a failure in the case where error information is contained in the Bayesian network of FIG. 4, and when error information is removed.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Image forming apparatus 11 Image reading part 12 Print engine part 13 Sensor part 14 Diagnosis information input part 15 Failure diagnosis part 16 Communication part 21 Component state information acquisition part 22 History information acquisition part 23 Environmental information acquisition part 24 Failure information acquisition part 25 Addition Operation information acquisition unit 26 Failure probability inference unit 27 Diagnosis result notification unit 28 Inference engine 29 Failure candidate extraction unit 30 Management device 31 Diagnostic model storage unit 32 Update information management unit 33 Probability update processing unit 34 Diagnostic model update unit 35 Communication unit 40 Market Quality information database 50 Market quality information input device

Claims (4)

Accumulation means for accumulating the number of occurrences of each failure cause, failure location, failure state, and treatment content related to each failure cause for each model of the image forming apparatus,
An acquisition unit that acquires, from the storage unit, the number of occurrences of all combinations of a failure part related to the failure cause and at least one of a failure phenomenon, a failure state, and a treatment content related to the failure cause for each failure cause; ,
Based on the number of occurrences for each cause of failure acquired by the acquisition means, an occurrence probability calculation means for calculating an occurrence probability between each node of the Bayesian network representing a failure diagnosis model of parts constituting the image forming apparatus,
Fault diagnosis system with   The occurrence probability calculation means multiplies the number of occurrences of all combinations acquired by the acquisition means by a weighting factor indicating the degree of relevance of each of the failure phenomenon, the failure site, the failure state, and the action content for the failure cause. The fault diagnosis system according to claim 1, further comprising: means; and sum calculation means for calculating the sum of the number of occurrences of all combinations of the results multiplied by the multiplication means. The acquisition means acquires the number of occurrences of all combinations further including a failure part of another image forming apparatus used before the image forming apparatus to be diagnosed,
3. The multiplication unit according to claim 2, wherein the multiplication unit further multiplies the number of occurrences of all combinations acquired by the acquisition unit by a weighting factor indicating a degree of association with the failure part of the other image forming apparatus with respect to the failure cause. Fault diagnosis system.   The fault diagnosis system according to any one of claims 1 to 3, further comprising inference means for inferring a cause of failure of the component based on the occurrence probability calculated by the occurrence probability calculation means.