JP2010114701A - Set data discrimination method, failure prediction method, set data discriminating device, failure predicting device, and image forming apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a failure predicting device 120 that accurately discriminates between the set data acquired in a normal state which is close to abnormality and the set data acquired in an abnormal state which is close to normality. <P>SOLUTION: The failure predicting device includes a set data group constructing means 121 and an algorithm constructing means 122. The set data group constructing means: acquires set data from a color image forming apparatus 1 in a prescribed period, stores the set data and constructs the time series data of the set data; sorts each of the acquired set data, based on which the cluster among a plurality of clusters the each of the acquired set data corresponds to by a cluster analysis; specifies a point of time, when an appearance pattern of a cluster starts to change, on the basis of the result of the sorting; and discriminates the set data acquired between the specified point of time and immediately prior to the occurrence of a failure as abnormal set data, while discriminating the set data acquired between a period of time of starting to acquire data and immediately prior to the appearance pattern starts to change as normal set data. The algorithm-constructing means constructs an algorithm for failure prediction, on the basis of the discrimination results. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  Whether the present invention is normal set data or abnormal set data about the set data consisting of a combination of a plurality of types of information acquired from the inspection machine equipped with a part or internal device to be subject to failure prediction The present invention relates to a group data sorting method and a group data sorting device for sorting. The present invention also relates to a failure prediction method, a failure prediction apparatus, and an image forming apparatus that are used for failure prediction by constructing a normal data group based on the normal set data sorted by the set data sorting method and the set data sorting device.

  Failures that occur in various devices on the market can be broadly divided into those that make it impossible to use the device and those that can cause the device to be usable but difficult to withstand. Is done. When the former failure occurs, the device must be not used until the repair is performed, so that a quick repair is required. Further, when the latter failure occurs, the device can be used, but causes a problem that it is difficult to withstand. For example, in an electrophotographic image forming apparatus, an abnormal image such as a black stripe may be formed due to a failure of a photoconductor as a component or a photoconductor cleaning device as an internal device. . Since the printing operation can be performed normally, an image can be formed, but there is a problem that it is difficult to withstand the use that only an abnormal image can be obtained. In the situation where such a problem occurs, in most cases, prompt repair is required. In other words, no matter what kind of failure occurs, prompt repair is required.

  However, it takes time to arrange parts and internal devices, and it is often inconvenient for a service person, so that it takes a certain period of time to carry out the repair.

  Therefore, conventionally, an electrophotographic image forming apparatus that prompts the user to periodically replace parts and internal devices is known. That is, for each part or internal device, the cumulative number of prints is counted individually, and when a part or internal device whose count value exceeds a predetermined threshold occurs, the user is prompted to replace that part or internal device. It is. As described above, by performing an operation of exchanging various parts and internal devices before the occurrence of a failure, it is possible to suppress the occurrence of a situation in which the user is inconvenienced until the repair is performed.

  However, since there are individual differences in the deterioration progress rate of parts and internal devices, in order to prompt the user to replace parts and internal devices before a failure occurs, the threshold value of the count value is set to a relatively low deterioration progress rate. It must be set according to the fast individual. In such a setting, although there is still a considerable margin before the occurrence of a failure, there are almost all cases in which replacement is promoted, so that a lot of useless costs are generated.

  On the other hand, recent devices often employ a configuration in which control parameters are adjusted based on detection results (hereinafter referred to as sensing data) by various sensors. And by combining those sensing data, there are many cases that can be used for failure prediction of components and internal devices. For example, in an electrophotographic image forming apparatus, an optical sensor that detects a toner adhesion amount per unit area with respect to a reference toner image, a toner concentration sensor that detects a toner density of a developer, and a potential that detects a surface potential of a photoreceptor. Various sensors such as sensors are installed. The present applicant previously disclosed a method for predicting the occurrence of a failure in a part of an image forming apparatus or an internal apparatus based on a result of processing sensing data from various sensors by an MTS (Maharanobis Taguchi System) method in Patent Document 1. is doing. In this method, first, a large amount of set data obtained by combining various types of sensing data is collected from an inspection machine in which components and internal devices are in a normal state to obtain a normal set data group. Thereafter, the inspection machine (the inspection machine that acquired the normal data group or another inspection machine having the same specifications) that is the target of failure prediction is operated. Then, the Mahalanobis distance indicating the relative positional relationship in the multidimensional space of the above-described normal set data group is obtained for the set data acquired from the operating test machine. The Mahalanobis distance of the set data gradually increases as the parts and internal devices of the test machine gradually approach from the normal state to the abnormal state (a state where a failure will occur soon). For this reason, a state where the Mahalanobis distance has reached a predetermined threshold value can be regarded as an abnormal state in which a failure occurs in a component or internal device in the near future. By prompting the user to replace the parts and the internal device at that time, it is possible to reduce generation of useless costs due to prompting the replacement even though it can still withstand use.

  In the method described in Patent Document 1, an abnormal state of a component or an internal device is detected by data processing using the MTS method. However, the abnormal state can also be detected by another method. For example, it is possible to detect an abnormal state of a component or an internal device by a well-known linear discriminant method or AdaBoost method (weighted majority vote discriminant method). These methods are different from the MTS method in that the normal group data group is not used for failure prediction as it is, but is used to construct an algorithm for failure prediction. On the other hand, a set of data consisting of a plurality of types of data is acquired from an inspection machine equipped with a part to be predicted for failure or an internal device, and a failure is predicted based on the result of processing with a predetermined algorithm. This is common with the MTS method. Even if we focus only on one type of data, it is not possible to determine with high accuracy whether or not it is an abnormal state, but it is possible to determine with high accuracy by combining multiple types of data. is there.

JP 2005-266380 A

  However, in any of the MTS method, the linear discriminating method, and the AdaBoost method, for the reason described below, an erroneous determination that is considered to be an abnormal state despite the normal state is likely to occur. That is, when a component or internal device is in a normal state or in an abnormal state that is a stage before a failure, a phenomenon that reveals a mechanical or functional failure such as an abnormal image or abnormal noise does not occur. For this reason, it is very difficult to discriminate between a normal state close to abnormal and an abnormal state close to normal. For these reasons, conventionally, only the set data acquired when it is clear that it is normal, such as during initial operation, that is, when the degree of normality is relatively large, is included in the normal set data group. I could not. This has been a cause of facilitating misjudgment that is considered to be an abnormal state despite the normal state.

  The reason why only the group data having a relatively high degree of normality is included in the normal group data group causes the above-described erroneous determination will be described by taking the MTS method as an example. The threshold value of the Mahalanobis distance is generally set to a value corresponding to the outside of the multi-dimensional space of the normal set data group. This is because if a value corresponding to the inner side of the multidimensional space is set, an erroneous determination that changes the normal state to the abnormal state occurs. In addition, with respect to the threshold, with the aim of accurately detecting an abnormality close to normal as an “abnormal state”, set data acquired in an abnormal state close to normal (hereinafter referred to as abnormal set data close to normal) It is desirable to set the same value. If abnormal group data close to normal can be included in the normal group data group, such setting can be easily realized by setting the threshold value to a value slightly larger than the spread of the multi-dimensional space of the normal group data group. it can. However, in a normal group data group that includes only a relatively large set of normality data, the Mahalanobis distance of the abnormal group data that is close to normal, that is, the appropriate threshold value is the spread of the multidimensional space of the normal group data group. Will be much larger than that. Since it is difficult to accurately grasp how large the value is, a large error is often generated between the actually set threshold value and the appropriate value. Due to the error, the Mahalanobis distance exceeds the threshold even though it is normal, causing an erroneous determination that it is abnormal.

  Further, in the linear discriminant method and the AdaBoost method, an abnormal group data group including only the group data acquired from the test machine in which the parts and internal devices are in an abnormal state is used in addition to the normal group data group. In such a method, in order to suppress the occurrence of a misjudgment that is considered to be a normal state in spite of being in an abnormal state, it is acquired when the component or internal device is in an abnormal state close to normal. It is necessary to include the set data in the abnormal set data group. However, conventionally, since it is difficult to discriminate between a normal state close to abnormal and an abnormal state close to normal, set data acquired immediately before the occurrence of a failure, that is, set data having a relatively high degree of abnormality However, it could not be included in the abnormal group data group. This has been an obstacle to improving the accuracy of failure prediction.

  The present invention has been made in view of the above background, and a first object of the present invention is to provide the following set data sorting method and set data device. In other words, with respect to the set data acquired from the machine under test, the set data acquired when the part or internal device is in a normal state close to an abnormality and the set data acquired when the part or internal device is in an abnormal state close to normal. A group data separation method that can accurately classify the set data.

  A second object is to provide the following failure prediction method, failure prediction apparatus, and image forming apparatus. That is, the failure prediction accuracy is deteriorated by erroneously determining that the component or internal device is in an abnormal state despite being in a normal state, or erroneously determining that it is in a normal state in spite of being in an abnormal state. It is a failure prediction method that can be suppressed.

In order to achieve the first object, the invention according to claim 1 is directed to a set data consisting of a combination of a plurality of types of information acquired from a machine equipped with a component or an internal device. In a group data classification method for classifying whether it is normal group data acquired when in a normal state or abnormal group data acquired when the part or internal device is in an abnormal state, at least, In a period from a predetermined start time to a time when a failure of the component or the internal device occurs, the set data is acquired and stored from the test machine at a predetermined cycle, and time series data of the set data is constructed. A data construction process and a classifier that classifies by cluster analysis which of the plurality of sets of data in the time-series data corresponds to each of a plurality of clusters. And, based on the classification result, specify the point in time before the point of failure occurrence and the point in time when the appearance pattern of the cluster starts to change as the transition point when the part or internal device has transitioned from the normal state to the abnormal state Among the plurality of set data in the process and the time series data, the set data acquired from the transition time to immediately before the failure occurrence time is classified as abnormal set data, while the transition time from the start time to the transition time And a separation step of separating the group data acquired until immediately before as normal group data.
In order to achieve the second object, the invention according to claim 2 is based on set data composed of a combination of a plurality of types of information acquired from a test machine equipped with components or internal devices. A normal group data group construction process for constructing a normal group data group consisting only of the group data, the group data acquired during operation for the inspection machine equipped with the part or internal device to be subject to failure prediction, and the In a failure prediction method for performing a prediction step of predicting the occurrence of a failure of the component or internal device based on the normal set data group, a plurality of pieces acquired from the test machine in the normal set data group construction step Concerning the group data, constructing a normal group data group composed only of normal group data based on the result of classifying whether the group data is normal group data or abnormal group data by the group data classification method of claim 1 It is an butterfly.
Further, the invention of claim 3 is based on a set data consisting of a combination of a plurality of types of information acquired from a machine equipped with a component or an internal device, and a normal set data group consisting of only the normal set data; A data group construction process for constructing an abnormal group data group consisting of only the abnormal group data, and a test machine equipped with the part or internal device to be subject to failure prediction, to the group data acquired during operation Based on the normal group data group and the abnormal group data group, an algorithm construction step for constructing an algorithm for predicting the occurrence of a failure in the part or the internal device based on the part or the internal device to be subject to failure prediction A prediction process for predicting the occurrence of a failure of the component or internal device based on the result of processing the set data acquired during operation with the algorithm In the failure prediction method, the plurality of set data acquired from the test machine in the set data group construction step is classified as normal set data or abnormal set data by the set data classification method according to claim 1. It is characterized by doing.
In order to achieve the first object, the invention of claim 4 relates to a set data consisting of a combination of a plurality of types of information acquired from a machine equipped with a component or an internal device. In a group data sorting device that sorts whether normal set data acquired when the device is in a normal state or abnormal set data acquired when the part or internal device is in an abnormal state, At least during a period from a predetermined start time to a time when a failure of the component or internal device occurs, the set data is acquired and stored at a predetermined cycle from the test machine, and time-series data of the set data is obtained. The data construction means to be constructed and the plurality of sets of data in the time-series data are classified by cluster analysis as to which of the plurality of clusters corresponds. Based on the classification means and the classification result, the point in time before the point of failure occurrence and the time when the appearance pattern of the cluster starts to change is identified as the transition point when the part or internal device has transitioned from the normal state to the abnormal state And identifying the group data acquired from the transition time to immediately before the failure occurrence time as abnormal group data among the plurality of the group data in the time-series data, while the transition from the start time The present invention is characterized in that there is provided sorting means for sorting the set data acquired immediately before the time point as normal set data.
In order to achieve the second object, the invention according to claim 5 is based on set data composed of a combination of a plurality of types of information obtained from a test machine equipped with components or internal devices. A normal set data group constructing means for constructing a normal set data group consisting only of set data, and the test data equipped with the parts or internal devices to be subject to failure prediction, the set data acquired during operation, In a failure prediction device comprising a prediction means for predicting the occurrence of a failure of the component or internal device based on a normal set data group, the set data classification device according to claim 4 is provided, and a plurality of pieces of data acquired from a test machine For the group data, the normal group data group consisting of only the normal group data is constructed based on the result of classifying whether the group data is normal group data or abnormal group data by the group data sorting device. It is characterized in that so as to build a stage.
Further, the invention of claim 6 is based on set data consisting of a combination of a plurality of types of information acquired from a test machine equipped with a component or internal device, and a normal set data group consisting of only the normal set data; Data group construction means for constructing an abnormal group data group consisting only of the abnormal group data, and a test machine equipped with the part or internal device to be subject to failure prediction, the group data acquired during operation Based on the normal group data group and the abnormal group data group, an algorithm construction unit for constructing an algorithm for predicting the occurrence of a failure of the part or the internal device based on the part or the internal device to be subject to failure prediction And a predicting means for predicting the occurrence of a failure of the component or the internal device based on the result of processing the set data acquired during operation with the algorithm. In the failure prediction apparatus, the set data classification device according to claim 4 is provided, and the plurality of set data acquired from the test machine is classified as normal set data or abnormal set data by the set data sorting device. On the basis of the above, the data group construction means is configured to construct a normal group data group consisting only of normal group data and an abnormal group data group consisting only of abnormal group data.
According to a seventh aspect of the present invention, there is provided an image forming apparatus comprising: an image forming unit that forms an image; and a failure predicting unit that predicts a failure of a component or an internal device. It is characterized by using a prediction device.

The inventors of the present invention have completed these inventions through the process described below. That is, as described above, when a component or internal device is in a normal state or in an abnormal state, a phenomenon that reveals a mechanical or functional defect such as an abnormal image or abnormal noise does not occur. For this reason, it is not possible to discriminate between a normal state close to abnormal and an abnormal state close to normal based on such a phenomenon. However, in the vicinity of the boundary between the normal state and the abnormal state, there is a possibility that some characteristic change is suddenly occurring in the set data. The present inventors paid attention to the behavior of the set data as the characteristic change. We thought that if the behavior of the unique set data was recognized during the abnormal period, it would be possible to identify the boundary between the two states based on the behavior. However, the group data includes a plurality of types of data, and even if the behavior of each data is viewed individually, it is impossible to grasp the behavior of each unit of the group data. Therefore, an attempt was made to grasp the behavior of each group by cluster analysis. As is well known, cluster analysis is an analysis technique for grouping individuals or variables that are similar to each other into a group (same cluster). If cluster analysis is used, it is possible to represent pair data having characteristics similar to each other by the same cluster, and to understand the behavior of the pair data of the pair data. Although the details of the experiment will be described later, a phenomenon has been confirmed in which the appearance pattern of the cluster of the set data starts to change abruptly at a point more than one week after the failure of the component or internal device. For example, during the period from the start of continuous printing until reaching a point one week before the occurrence of the failure, there were mainly three types of clusters of A, B, and C. When it reaches, it changes to a pattern in which three types of clusters B, C, and D appear, and thereafter, a pattern in which three types of clusters C, D, and E appear, and four types of patterns D, E, F, and G The appearance pattern changes sequentially until the failure occurs, such as a pattern where a cluster appears. Under the conditions in which such a phenomenon occurs, the time point when the appearance pattern of the cluster starts to change can be regarded as the transition time point when the component or the internal device has shifted from the normal state to the abnormal state.
In the group data classification method according to claim 1 and the group data classification device according to claim 4, by specifying the time of the transition, the parts and internal devices of the group data acquired from the machine to be tested are normal. It is possible to accurately distinguish the set data acquired when the state is in the state and the set data acquired when the component or the internal device is in an abnormal state close to normal.
In the invention of the failure prediction method, failure prediction apparatus, or image forming apparatus, the set data classification method according to claim 1 is used as the set data classification method, so that parts and internal devices are in a normal state close to abnormal. The group data acquired sometimes is included in the normal group data group, or the group data acquired when the component or the internal device is in an abnormal state close to normal is included in the abnormal group data group. By using such a normal group data group and an abnormal group data group, it is assumed that the normal state is in spite of being in an abnormal state, or being in a normal state in spite of being in a normal state. The occurrence of erroneous determination can be suppressed.

First, a color image forming apparatus will be described as an example of a test machine before an embodiment of a group data sorting method to which the present invention is applied.
FIG. 1 is a schematic configuration diagram illustrating a color image forming apparatus 1 as an example of a test machine. FIG. 2 is a block diagram illustrating a connection state between the control unit 100 mounted on the color image forming apparatus 1 and the surrounding electrical devices.

  In FIG. 1, the color image forming apparatus 1 includes a paper feeding unit 10, a transfer unit 20 having an intermediate transfer belt 21, and yellow (Y), magenta (M), cyan (colored) disposed along the intermediate transfer belt 21. The image forming units 30Y, 30M, 30C, and 30Bk, which are toner image forming units for the respective colors C) and black (Bk), are provided in a main body housing (not shown). Further, an adhesion amount detection unit 50 for detecting the toner adhesion amount of the toner image on the intermediate transfer belt 21 is provided. In addition to these, a control unit 100 for controlling each device of the color image forming apparatus 1 is provided.

  The image forming unit 30 for each color will be described. Here, the Bk color image forming unit 30Bk will be described, but the Y, M, and C image forming units 30Y to 30C have the same configuration. In the image forming unit 30Bk, a charging unit 32Bk, an exposure unit 33Bk, a developing unit 34Bk, a primary transfer unit 35Bk, a cleaning unit 36Bk, and the like are disposed around the photoreceptor 31Bk.

  At the time of image formation, when a normal operation signal is instructed by the controller 110 of the color image forming apparatus, the photoreceptor 31Bk is rotationally driven by a drive motor (not shown) under the control of the control unit 100. Further, as shown in FIG. 2, the CPU of the control unit 100 sequentially outputs a bias means for each image forming process including a driving means such as a photoconductor motor and a charging bias sequentially. The color image signal from the external device is subjected to image processing such as color conversion processing by the image signal generation circuit 101 of the control unit 100, and a Bk color image signal is output to the exposure unit 33Bk. The exposure unit 33Bk converts the Bk image signal into an optical signal by the exposure drive circuit 102 of the control unit 100, and scans and exposes the photoconductor 31Bk while the exposure laser diode blinks based on the optical signal. Thus, an electrostatic latent image is formed.

  The electrostatic latent image on the photoreceptor 31Bk is developed by the developing unit 34Bk to become a Bk toner image, and the Bk toner image on the photoreceptor 31Bk is transferred onto the intermediate transfer belt 21 by the transfer unit 35Bk. The photoreceptor 31Bk is prepared for the next image formation by cleaning the residual toner by the cleaning unit 36Bk after transferring the toner image and removing the charge by the charge removing lamp 38Bk.

  Similarly, the image forming units 30Y, 30M, and 30C include a charging unit, a developing unit, a cleaning unit, a charge removal lamp, and the like around the photoreceptors 31Y, 31M, and 31C. Then, Y, M, and C toner images are formed on the photoreceptors 31Y, 31M, and 31C, and these images are primarily transferred while being superimposed on the intermediate transfer belt 21.

  A transfer unit 20 serving as a transfer unit is disposed below the image forming unit for each color. The transfer unit 20 includes an endless intermediate transfer belt 21, driven rollers 22, 23, a driving roller 24, and the like. An intermediate transfer belt 21, which is an image carrier that carries a plurality of color toner images, is stretched around a drive roller 24, driven rollers 22, 23, and the like. The intermediate transfer belt 21 is rotationally driven counterclockwise in FIG. 1 by the drive roller 24 being rotationally driven by a drive mechanism such as a motor (not shown) under the control of the control unit 100 shown in FIG. The Y, M, C, and Bk toner images formed on the photoconductors 31Y, 31M, 31C, and 31BB for each color are primarily transferred onto the intermediate transfer belt 21 in a primary transfer nip for each color. As a result, a four-color superimposed toner image (hereinafter referred to as a four-color toner image) is formed on the intermediate transfer belt 21.

  A secondary transfer bias roller 61 is in contact with the driving roller 24 around the intermediate transfer belt 21 from the belt front surface side, thereby forming a secondary transfer nip 6. As shown in FIG. 2, a secondary transfer bias is applied to the secondary transfer bias roller 61 by a bias power supply circuit 104 under the control of the control unit 100. Thereby, a secondary transfer electric field is formed between the secondary transfer bias roller 61 and the grounded secondary transfer nip back roller 24. The four-color toner image formed on the intermediate transfer belt 21 enters the secondary transfer nip as the belt moves endlessly.

  The paper supply unit 10 separates the recording paper (transfer paper) 12 in the paper supply cassette 11 one by one by, for example, a paper supply roller 11a and a separation member 11b (not shown) and sends them to a pair of registration rollers (not shown). The registration roller pair adjusts the timing of the recording paper 12 sent from the paper feed cassette 11 and sends the recording paper 12 toward the secondary transfer nip 6 at a predetermined timing. At the secondary transfer nip 6, the four-color toner image on the intermediate transfer belt 21 is collectively transferred onto the recording paper 12 by the action of the secondary transfer electric field and nip pressure, and combined with the white color of the recording paper 12, Become.

  The recording paper 12 on which the full color image is formed in this way is conveyed to the fixing unit 40. The fixing unit 40 heats and presses the recording paper 12 on which the full-color image is formed by the fixing roller 41 and the pressure roller 42 to fix the toner of each color on the recording paper 12, and the discharge roller pair (not shown) discharges the toner. Eject onto the paper tray.

  In the image forming unit 30 including the photosensitive member 31, the charging unit 32, the exposure unit 33, the developing unit 34, the primary transfer unit 35, and the cleaning unit 36, the exposure unit 33 and the primary transfer among these units. The part excluding the part 35 is configured as a process unit. In this process unit, the photosensitive member 31, the charging unit 32, the developing unit 34, and the cleaning unit 36 are held by a common holding unit, and these are integrally attached to and detached from the image forming apparatus main body as one unit. Is.

  The adhesion amount detection unit 50 is disposed downstream of the black (Bk) image forming unit 30Bk in the movement direction of the intermediate transfer belt 21, and as shown in FIG. A pair of optical sensors 51 and 52 as optical detection means are provided. As shown in FIGS. 4 and 5, the optical sensors 51 and 52 are respectively a light emitting element 151 made of a light emitting diode, a first light receiving element 152 that receives diffusely reflected light, and a second light receiving element that receives regular reflected light. 153. The first light receiving element 152 and the second light receiving element 153 use Si phototransistors, PDs (photodiodes), or the like. Each element 151, 152, 153 is mounted on the printed circuit board 150. In addition, a condensing lens 154 is disposed on the exit optical path, and the exit light from the light emitting element 151 is refracted by the condensing lens 154 and condensed on the irradiation target on the surface of the intermediate transfer belt 21 as an image carrier. Is done. Also, condensing lenses 155 and 156 are arranged on the incident optical path. The light receiving elements 152 and 153 receive the light collected by the condensing lenses 155 and 156 from the reflected light reflected from the toner that is the irradiation object on the intermediate transfer belt 12. The printed circuit board 150 is connected to the control unit 100. A voltage adjusted by the light amount adjustment circuit 105 of the control unit 100 shown in FIG. 2 is applied to the light emitting element 151. In addition, the control unit 100 converts the output signals from the first and second light receiving elements 152 and 153 into digital signals by the AD converter 106.

  As the optical sensors 51 and 52, those capable of detecting near infrared light and / or infrared light are used. Near-infrared light and / or infrared light are not affected by the colorant of the toner if the toner adhesion amount of the toner image is the same, and the output values of the light receiving elements show substantially the same value. Specifically, an optical element that emits light having a peak emission wavelength of about 840 [nm] is used, and a light receiving element having a peak spectral sensitivity of about 840 [nm] is used. Further, for example, the light emitting element may be a light emitting element that emits light from visible light to infrared light, and the light receiving element may be a light receiving element that receives near infrared light or infrared light. The light receiving element may be a light receiving element that receives light in a region from visible light to infrared light, and the light emitting element may be a light emitting element that emits near infrared light or infrared light. Even if the optical sensor has such a configuration, it can be an optical sensor that detects near-infrared light or infrared light. Note that when low-cost carbon black is used as the colorant for the black toner, carbon exhibits strong light absorption even in the infrared region, and as shown in FIG. 6, the amount of adhesion is detected compared to Y, M, and C colors. Sensitivity is lowered.

  In the color image forming apparatus shown in FIG. 1, a process adjustment operation for adjusting a developing bias, a charging bias, an exposure amount and the like to optimize the image density of each color when the power is turned on or every time a predetermined number of prints are performed. Has been done. In the electrophotographic system, there is a weak point that the image density fluctuates due to deterioration with time and environmental fluctuations. Therefore, the process adjustment operation is executed to control the image density to be stable.

  The control flow of this process adjustment operation is shown in FIG. A process adjustment operation signal is instructed to the controller 100 by the controller 110 when the power is turned on or before and after printing a predetermined number of sheets, and the process adjustment operation starts (see FIG. 2).

  When the process adjustment operation is started, the control unit 100 sets the image signal generation circuit 101 to the image pear state (S201). Next, as shown in FIG. 4, the CPU of the control unit 100 irradiates the intermediate transfer belt 21 with light and receives the specularly reflected light with the second light receiving element 153. Then, the light intensity adjustment circuit 105 adjusts the light emission intensity R of the light emitting element 151 of the optical sensors 51 and 52 so that the output (light reception signal) of the second light receiving element 153 becomes a predetermined value (S202 to S204). ). As shown in FIG. 8, the output value of the second light receiving element 153 varies due to individual differences in light emission efficiency of the light emitting element 151, temperature variation, and variation with time. Therefore, the toner image density can be accurately measured by adjusting the emission intensity R of the light emitting element 151 so that the output value of the second light receiving element 153 becomes the target output value. That is, S202 to S204 correspond to the calibration operation of the optical sensors 51 and 52 for accurately measuring the adhesion amount of the toner image by the optical sensors 51 and 52.

  When the calibration operation of the optical sensors 51 and 52 is completed, a pattern image 60 as shown in FIG. 9 is automatically formed on the intermediate transfer belt 21 at a position facing the optical sensors 51 and 52 (S205). The pattern image 60 is composed of about five patch images 60S having different density levels, and includes a Bk pattern image 60Bk, an M pattern image 60M, a C pattern image 60C (not shown), and a Y pattern image. 60Y (not shown) is sequentially formed on the intermediate transfer belt 12. The patch image 60S is formed by changing the exposure conditions. At this time, the charging and developing bias conditions are executed with predetermined specific values. The pattern image on the intermediate transfer belt is optically measured by the optical sensors 51 and 52 as shown in FIG. 5 (S206).

  Next, the five light reception signals of the first light receiving element 152 that receives the irregularly reflected light obtained by detecting each patch image 60S of each color pattern image are used as the adhesion amount and the output of the light receiving element as shown in FIG. Conversion processing into toner adhesion amount (image density) is performed using an adhesion amount calculation algorithm constructed based on the relationship with the value. Thereby, the toner adhesion amount of each patch image 60S is detected. In the color image forming apparatus shown in FIG. 1, an optical sensor using near infrared and / or infrared light is used, and there is no difference in the output value of the first light receiving element 152 depending on the color. There is no need to provide an algorithm for each color, and a common adhesion amount calculation algorithm can be used. When carbon black is used as the black colorant, as shown in FIG. 6, the output values of the light receiving elements with respect to the adhesion amount are different between Y, M, C, and Bk. Two adhesion amount calculation algorithms for M and C and for Bk are used.

  When the toner adhesion amount of each patch image 60S is detected for each color, the relationship between the toner adhesion amount (per unit area) of each patch image and each development potential when each patch image is created is as shown in FIG. Then, a linearly approximated development potential-toner adhesion amount straight line is obtained for each color. The inclination γ and the intercept x0 are calculated for each color from the development potential-toner adhesion amount straight line (S207). By obtaining the slope γ and intercept x0 of each color in this way, the slope γ and intercept x0 of the straight line are deviated from the target characteristics (dotted line in the figure) due to the concentration variation factors (deterioration with time / environmental variation) described above. Can be detected. An exposure light amount correction parameter P for correcting the deviation of the inclination γ is determined from the inclination γ. Further, a development bias correction parameter Q is determined from the intercept x0 in order to correct a deviation of the development potential (intercept X0) at which development is started (S208).

  By multiplying the exposure light amount correction parameter P by the exposure signal, the inclination γ is mainly corrected, and by multiplying the development bias by the correction parameter Q, the intercept x0 is mainly corrected, thereby stabilizing the target image density. Can be obtained. In the above description, the exposure light amount and the development bias are corrected. However, other process control values that contribute to the image density such as a charging potential and a transfer current may be corrected.

  The cleaning units 36Y, 36M, 36C, 36Bk for cleaning the surface of the photoconductor in the image forming units 30Y, 30M, 30C, and 30Bk of the respective colors are not transferred to the intermediate transfer belt 21 as follows, , C and Bk, the transfer residual toner remaining on the surface is scraped off. When the blade member BL is worn due to long-term use, the scraping force of the blade member BL is reduced. As a result, a black streak image may occur. A black streak image may also be generated when the photoconductors 31Y, 31M, 31C, and 31B cause a failure due to deterioration.

Next, experiments conducted by the present inventors will be described.
The inventors prepared a printer testing machine having the same configuration as that of the color image forming apparatus 1 shown in FIG. Assuming that each color process unit, which is an internal device of this printer testing machine, is the target of failure prediction, a process unit for Bk that has confirmed that no initial failure has been caused to the printer testing machine. equipped. Then, continuous printing for continuously outputting a predetermined monochrome test image was performed. During the continuous printing, every time 100 sheets are output, the print job is temporarily stopped, and the above-described process adjustment operation is performed for the Bk process unit. Then, exposure light amount correction parameter P (hereinafter referred to as P value), development bias correction parameter Q (hereinafter referred to as Q value), and emission intensity R (hereinafter referred to as R value) for the Bk process unit obtained by the process adjustment operation. Etc. were sequentially stored in the data storage means. Then, at least during the period from the start of continuous printing to the time when a black streak image, which is an abnormal image, is generated (at the time when a failure occurs), set data consisting of data such as P value, Q value, R value, etc. is sequentially acquired. And save and build time series data.

  Thereafter, a plurality of sets of data included in the obtained time series data were classified into clusters by a well-known cluster analysis method. Then, the behavior of the cluster of the set data in the time series indicated by the time series data was analyzed. Specifically, the period from the start of continuous printing to the point at which the black streak image that is an abnormal image is generated is divided into predetermined unit times, and the period from the start to the end of the period is divided into a plurality of elapsed time zones. It was. Then, in each elapsed time zone, we analyzed what pattern cluster data appeared in. Then, from the start of continuous printing to the point of reaching one week or more before the occurrence of the failure, the specific appearance pattern continued all the time, but the phenomenon that the pattern started to change suddenly at the same point was confirmed. For example, a pattern in which three types of clusters A, B, and C appear until the simultaneous point is reached, whereas a pattern in which three types of clusters B, C, and D appear when the point is reached Then, a pattern in which three types of clusters C, D, and E appear, a pattern in which four types of clusters D, E, F, and G appear, and so on. When the experiment was repeated, a relatively high reproducibility was recognized in that the appearance pattern started to change. Therefore, the time point at which the appearance pattern of the cluster begins to change can be regarded as the transition time point when the Bk process unit, which is an internal component, has shifted from the normal state to the abnormal state.

  Although it has been experimentally verified that there is a change in the appearance pattern of clusters that occurs when a process unit transitions from a normal state to an abnormal state, the same phenomenon occurs not only in the process unit but also in various equipment parts and internal devices. The likelihood of being recognized is considered very high.

Next, an embodiment of a failure prediction apparatus to which the present invention is applied will be described.
The failure prediction apparatus according to the embodiment predicts failures of various components and internal devices mounted on the color image forming apparatus 1 that is a test machine shown in FIG. More specifically, at least for each color process unit as an internal device, the occurrence of a failure is individually predicted. In addition to or instead of the process units for each color, internal devices such as the transfer unit 20 and the exposure unit may be targeted for failure prediction. In addition, parts such as the intermediate transfer belt 21 may be targeted for failure prediction.

  The failure prediction apparatus according to the embodiment includes at least a set data group construction unit and a prediction unit. Prior to performing failure prediction, the set data group constructing means acquires set data (including at least P value, Q value, and R value) from the color image forming apparatus 1 to obtain a normal set. Data group and abnormal group data group are constructed. Further, the predicting means is based on the set data acquired during operation and the normal set data group that has been built in advance for the color image forming apparatus 1 on which the parts and internal devices that are the targets of failure prediction are mounted. The occurrence of a failure is individually predicted for at least each color process unit.

  The group data group construction means includes group data classification means, and constructs a normal group data group and an abnormal group data group based on the group data classification result by the group data classification means.

  The group data classification means provided in the group data group construction means includes data construction means, classification means, identification means, and classification means. The data construction means individually acquires and stores set data unique to each color at a predetermined cycle for each process unit of each color at least during a period from a predetermined start time to a time when a failure occurs. A data construction process (data construction process) for constructing the time-series data is performed. For example, for a Bk process unit, in order to predict its failure individually, it consists of a P value specific to Bk, etc., from a predetermined start time until a failure of the Bk process unit occurs. The set data is acquired and stored, and Bk time-series data is constructed.

  As for the predetermined start time, the initial operation after shipment from the factory, the start of operation after repairing parts or internal devices, etc. may be adopted.

  Further, the point in time when the failure starts is specified based on the abnormality occurrence information for each color process unit input from the operator. However, since it is not necessary to accurately grasp the point in time when the failure started, the acquisition of the set data and the time series based on the repair history information (repair execution information) input by the operator (user or serviceman) Data construction may be terminated. This is because a failure always occurs when repairs are performed. According to the experiments by the inventors, when the appearance pattern of a cluster starts to change due to the abnormal state of a process unit, the change continues even after a failure occurs until the unit is finally repaired. Followed. Therefore, it is possible to accurately specify the transition point from the normal state to the abnormal state based on the change in the appearance pattern of the cluster without accurately grasping the time point when the failure occurs.

  The classification means implements a classification process (classification process) for classifying, by cluster analysis, each of a plurality of set data in the time series data constructed by the set data group construction means. To do. This classification process is performed individually for each color.

  In addition, the identification unit determines whether the process unit is changed from a normal state to an abnormal state at a time point before the time when the failure of the process unit occurs and the appearance pattern of the cluster starts to change based on the classification result by the classification unit. The specific process (specific process) specified as the transition point of transition is performed. This specifying process is performed individually for each color.

  Further, the sorting means sorts the set data acquired from the transition time to immediately before the failure occurrence time among the plurality of set data included in the time series data as abnormal set data. Further, the group data acquired from the start time to immediately before the transition time is sorted as normal group data. Such a separation process (sorting step) is performed individually for each color.

  The group data group construction means of the failure prediction device constructs the normal group data group and the abnormal group data group by combining only the normal group data sorted by the sorting means in this way or by combining only the abnormal group data. The group data group construction process (group data group construction process) is performed. This set data group construction processing is also performed for each color individually.

  When a normal set data group for each color process unit is constructed in this way, the prediction means of the failure prediction apparatus starts failure prediction for each color process unit. Specifically, set data is acquired from the color image forming apparatus 1 at a predetermined cycle, and a Mahalanobis distance with respect to a normal set data group constructed in advance in the acquisition result is obtained. Then, the Mahalanobis distance is compared with a predetermined threshold value, and when the Mahalanobis distance is equal to or greater than the threshold value, it is predicted that a failure will occur soon, and this is notified to the user by a display display or the like. When the Mahalanobis distance is less than the threshold value, it is assumed that no failure has occurred yet, and the above notification is not performed. That is, the failure prediction apparatus according to the embodiment performs prediction processing using the MTS method. Such a prediction process (prediction process) is performed individually for each color process unit.

  In the failure prediction apparatus having the above-described configuration, the process data of each color is included when the set data acquired when it is in a normal state close to an abnormality is included in the normal set data group or acquired when the abnormal state is close to normal. The set data is included in the abnormal group data group. And by using such a normal group data group and an abnormal group data group, it is a misjudgment that it is in an abnormal state despite being in a normal state, and it is in a normal state in spite of being in an abnormal state It is possible to suppress the occurrence of misjudgment.

  The failure prediction apparatus according to the embodiment can be installed in the color image forming apparatus 1. For example, FIG. 2 described above shows an example of the color image forming apparatus 1 equipped with the failure prediction apparatus 120 according to the embodiment.

  Next, a modification of the failure prediction apparatus 120 according to the embodiment will be described. Unless otherwise specified below, the configuration of the failure prediction apparatus 120 according to the modification is the same as that of the embodiment.

  FIG. 11 is a block diagram illustrating the failure prediction apparatus 120 according to the modification together with the color image forming apparatus 1 and the like. In the figure, the failure prediction apparatus 120 has an algorithm construction means 122 in addition to the set data construction means 121 and the prediction means 123. The group data construction unit 121 constructs a normal group data group and an abnormal group data group individually for each color process unit in the same manner as in the embodiment.

  The algorithm construction means 122 constructs an algorithm for constructing an algorithm for predicting a process unit failure based on the set data acquired from the color image forming apparatus 1 by a known linear discriminant method, AdaBoost method, or principal component analysis method. Perform processing (algorithm construction process). This algorithm construction process is also performed individually for each color process unit.

  The algorithm construction means 122 constructs an algorithm as follows when the linear discriminant method is used. That is, first, for each color process unit, a linear discriminant function for discriminating between a normal state and an abnormal state based on the normal group data group and the abnormal group data group is constructed as the aforementioned algorithm. In this case, the prediction unit 123 detects an abnormal state based on the result of processing the set data acquired from the color image forming apparatus 1 for each color process unit with the above-described linear discriminant function.

  Further, when the AdaBoost method is used, the algorithm construction unit 122 constructs an algorithm as follows. That is, the normal group data group and the abnormal group data group are processed by a supervised learning algorithm called a boosting method (see “Statistical Pattern Identification Information Geometry”, mathematical science, No. 489, MARCH, 2004). As a result, a numerical weighting coefficient corresponding to the degree of contribution to the detection of the abnormal state of the process unit is obtained as the above-described algorithm for each of various data included in the set data. In this case, the prediction unit 123 individually predicts the failure of each color process unit as follows. That is, the small discrimination process is performed on the data in each place included in the set data to individually determine whether or not the data is normal. In each small discrimination process, the discrimination result is expressed by a weighting coefficient specific to the data type. At this time, when the determination result is normal, the polarity of the weighting coefficient is made positive (or negative), while when it is abnormal, the polarity is reversed. The weighting coefficient is a numerical value that is individually set for each type of data. The larger the weighting coefficient is set for the data that greatly contributes to the detection of the abnormal state of the process unit by the boosting method. When small discrimination processing is performed for all types of data included in the set data, a total value obtained by summing up the results is calculated. Depending on whether the polarity of the total value is positive or negative, a final determination is made as to whether or not the component or internal device is normal. That is, when the polarity of the total value becomes positive or negative, an abnormal state of the process unit is detected, and a signal indicating that a failure is about to occur is output.

  Further, when using the principal component analysis method, the algorithm construction means 122 constructs an algorithm as follows. That is, principal component analysis is performed on the normal group data group and the abnormal group data group, and formulas for calculating the principal component scores from the primary to a predetermined dimension are constructed as algorithms. In this case, the prediction unit 123 detects an abnormal state with the process uni of each color based on the result obtained by processing the set data acquired from the color image forming apparatus 1 using the above-described formula.

  The example in which the color image forming apparatus 1 that acquires the normal group data group and the abnormal group data group is the same as the color image forming apparatus 1 that is the target of failure prediction of the process unit has been described. For example, they may be different. In this case, for example, the failure prediction apparatus 120 is connected to each of a plurality of color image forming apparatuses via a telephone line or the like, and normal group data groups and abnormal group data are obtained based on group data acquired from the respective color image forming apparatuses You may make it build a group. Further, in this case, it is only necessary to grasp the point in time when the failure has started by communication from the user or input operation to the color image forming apparatus. Further, instead of grasping when the failure has started, the timing at which the repair is performed may be grasped by contact from the user or a service person or input operation.

  Further, based on the constructed normal group data group and abnormal group data group and the group data individually acquired from each user's color image forming apparatus through a telephone line or the like, the process unit of each user's color image forming apparatus The failure prediction for each may be performed individually.

1 is a schematic configuration diagram showing a color image forming apparatus as a test machine. FIG. 3 is a block diagram illustrating a connection state between a control unit mounted on the color image forming apparatus and electric devices around the control unit. FIG. 3 is a perspective view showing four photoconductors and a transfer unit of the color image forming apparatus. The schematic diagram which shows the adhesion amount detection part which makes the test object the solid surface of an intermediate transfer belt. FIG. 3 is a schematic diagram illustrating an adhesion amount detection unit that uses a toner image on an intermediate transfer belt as a test target. The graph which shows the relationship between the output from the light receiving element which receives irregularly reflected light, and a toner adhesion amount. 6 is a flowchart showing a control flow of a process adjustment operation in the color image forming apparatus. The graph which shows the relationship between the output from the light receiving element which receives regular reflection light, and LED electric current (light emission amount). FIG. 3 is a schematic diagram showing a pattern image formed on the surface of an intermediate transfer belt. 6 is a graph showing the relationship between the toner adhesion amount and the development potential for a pattern image. The block diagram which shows the failure prediction apparatus which concerns on a modification with the same color image forming apparatus.

Explanation of symbols

1: Color image forming device (test machine)
120: Failure prediction device 121: Set data group construction means 122: Algorithm construction means 123: Prediction means

Claims (7)

About the set data consisting of a combination of a plurality of types of information acquired from a test machine equipped with a part or internal device, is the normal set data acquired when the part or internal device is in a normal state, or In a group data classification method for classifying whether the part or internal device is abnormal group data acquired when it is in an abnormal state,
At least during a period from a predetermined start time to a time when a failure of the component or internal device occurs, the set data is acquired and stored at a predetermined cycle from the test machine, and time-series data of the set data is obtained. A data construction process to construct;
A classification step of classifying which of a plurality of clusters corresponds to each of the plurality of sets of data in the time series data by cluster analysis;
A specific step of identifying, based on the classification result, a point in time before the occurrence of the failure and a point at which the appearance pattern of the cluster starts to change as a transition point at which the part or internal device has shifted from a normal state to an abnormal state; ,
Among the plurality of set data in the time series data, the set data acquired from the transition time to immediately before the failure occurrence time is classified as abnormal set data, while the start data is immediately before the transition time A group data classification method, comprising: performing a classification step of classifying the acquired group data as normal group data. A normal set data group construction step of building a normal set data group consisting only of the normal set data, based on the set data consisting of a combination of a plurality of types of information acquired from a test machine equipped with components or internal devices;
Predicting the occurrence of a failure of the component or internal device based on the set data acquired during operation and the normal set data group for the inspection machine equipped with the component or internal device subject to failure prediction In the failure prediction method for performing the prediction process
Based on the result obtained by classifying the plurality of set data acquired from the test machine in the normal set data group building process as normal set data or abnormal set data by the set data sorting method according to claim 1. A failure prediction method characterized by constructing a normal group data group consisting of only normal group data. Based on set data consisting of a combination of multiple types of information acquired from a test machine equipped with parts or internal devices, a normal set data group consisting only of the normal set data and an abnormality consisting only of the abnormal set data A data group construction process for constructing a tuple data group;
An algorithm for predicting the occurrence of a failure of the component or internal device based on the set data acquired during operation for the inspection machine equipped with the component or internal device subject to failure prediction, An algorithm construction process for construction based on the data group and the abnormal group data group;
Predicting the occurrence of failure of the component or internal device based on the result of processing the set data acquired during operation with the algorithm for the inspection machine equipped with the component or internal device subject to failure prediction In the failure prediction method for performing the prediction process
In the set data group construction step, the plurality of set data acquired from the test machine is classified as normal set data or abnormal set data by the set data classification method according to claim 1. Failure prediction method. About the set data consisting of a combination of a plurality of types of information acquired from a test machine equipped with a part or internal device, is the normal set data acquired when the part or internal device is in a normal state, or In the group data sorting device for sorting whether the part or the internal device is abnormal group data acquired when it is in an abnormal state,
At least during a period from a predetermined start time to a time when a failure of the component or internal device occurs, the set data is acquired and stored at a predetermined cycle from the test machine, and time-series data of the set data is obtained. Data construction means to construct;
Classifying means for classifying which of the plurality of clusters corresponds to each of the plurality of sets of data in the time series data by cluster analysis,
A specifying means for specifying, based on the classification result, a time point before the time of occurrence of the failure and a time point when the appearance pattern of the cluster starts to change as a transition time point when the component or internal device has shifted from a normal state to an abnormal state; ,
Among the plurality of set data in the time series data, the set data acquired from the transition time to immediately before the failure occurrence time is classified as abnormal set data, while the start data is immediately before the transition time A group data sorting device, comprising: a sorting unit that sorts the acquired group data as normal group data. Normal group data group construction means for constructing a normal group data group consisting only of the normal set data based on the group data consisting of a combination of a plurality of types of information acquired from a test machine equipped with parts or internal devices;
Predicting the occurrence of a failure of the component or internal device based on the set data acquired during operation and the normal set data group for the inspection machine equipped with the component or internal device subject to failure prediction In a failure prediction apparatus comprising a prediction means for
The set data sorting device according to claim 4 is provided, and a plurality of set data acquired from the test machine is determined based on the result of sorting whether the set data sorting device is normal set data or abnormal set data. A failure prediction apparatus characterized in that a normal group data group consisting only of group data is constructed by the normal group data group construction means. Based on set data consisting of a combination of multiple types of information acquired from a test machine equipped with parts or internal devices, a normal set data group consisting only of the normal set data and an abnormality consisting only of the abnormal set data A data group construction means for constructing a tuple data group;
An algorithm for predicting the occurrence of a failure of the component or internal device based on the set data acquired during operation for the inspection machine equipped with the component or internal device subject to failure prediction, Algorithm construction means for construction based on the data group and the abnormal group data group,
Predicting the occurrence of failure of the component or internal device based on the result of processing the set data acquired during operation with the algorithm for the inspection machine equipped with the component or internal device subject to failure prediction In a failure prediction apparatus comprising a prediction means for
The set data sorting device according to claim 4 is provided, and a plurality of set data acquired from the test machine is determined based on the result of sorting whether the set data sorting device is normal set data or abnormal set data. A failure prediction apparatus characterized by causing the data group construction means to construct a normal group data group consisting only of group data and an abnormal group data group consisting only of abnormal group data. In an image forming apparatus comprising: an image forming unit that forms an image; and a failure prediction unit that predicts a failure occurrence of a component or an internal device.
An image forming apparatus using the failure prediction apparatus according to claim 6 as the failure prediction means.

Images