JP5122254B2 - Operating state determination method and image forming apparatus - Google Patents

Description

  The present invention relates to an operation state determination method and an image forming apparatus, and more particularly to determination of a failure or a failure sign state in an apparatus such as an image forming apparatus in an operation state.

  In an image forming apparatus such as a copying machine, a printer, a facsimile machine, or a printing machine, for example, in the case of using an electrophotographic system, a static image corresponding to an original or image information is applied to a photosensitive member used as a latent image carrier. It is known that an electrostatic latent image is formed, and the electrostatic latent image is subjected to visible image processing with toner supplied from a developing device, and then the image transferred to a recording sheet or the like is fixed to produce a copy output. .

By the way, for example, an image forming apparatus provided with an image carrier such as a photosensitive member or an intermediate transfer belt is not only subjected to frictional wear associated with normal operation, but also contains harmful substances such as paper dust from the outside or unexpected operation. The function of the image forming apparatus is gradually deteriorated due to an increase in adhesive force due to excessive stirring of toner, dropping of external additives, wear of the cleaning means, contamination deterioration of the charging means, accidental failure, and the like, and the image forming apparatus enters a failure state. Such a failure results in a decrease in image quality. Specifically, it is an unpleasant abnormal image in the vertical direction along the direction of rotation, blurring of the image, an abnormal image in the shape of a stripe perpendicular to the rotation direction, a spot-like spot image or a white-out image. However, since there is no hindrance to the image forming operation of the device itself, it will continue to operate even if there is a malfunction, and when the user looks at the image, he notices the problem and repairs and forms the image. It is necessary to redo the process, and a large amount of time and resources are wasted.
In particular, since an electrophotographic image forming apparatus has a relatively complicated configuration and a large number of parts, it is likely that an abnormality will occur suddenly unless various parts are regularly maintained. .

As a technique for avoiding such a situation, a signal that predicts a failure is detected from signals from various sensors provided in the image forming apparatus (predictive detection), and a failure is predicted (failure prediction). Technology) has been proposed (for example, Patent Document 1).
Patent Document 1 discloses a configuration in which an index value is calculated from a plurality of types of information and a change in the state of the image forming apparatus is determined based on time change data of the index value.

In addition, a method (for example, Patent Document 2) for detecting a failure of a specific unit or component included in the apparatus, a method for predicting the lifetime of the unit or component (for example, Patent Documents 3 to 6), and A method for sequentially diagnosing various faults using these methods has been proposed (for example, Patent Document 7).
In addition to failure and failure prediction in units and parts, a method for detecting abnormality of a toner image obtained as a result of failure in these devices (for example, Patent Documents 8 to 10), and optically detecting the surface state of a photoreceptor An observation method has also been proposed (for example, Patent Documents 11 to 13).
In addition, a virtual sensor based on quantitative analysis using a hybrid diagnostic methodology is provided, and after formatting with an event generator, a method of performing fault diagnosis and predictive diagnosis by using a qualitative analysis device that is regarded as a diagnostic system for the hybrid is also proposed. (For example, Patent Document 14).

JP 2005-17874 A Japanese Patent Laid-Open No. 5-281809 Japanese Patent Laid-Open No. 5-100517 JP 7-36323 A JP-A-7-104616 JP 2001-356655 A JP 2000-270141 A JP-A-8-137344 JP 2000-89623 A JP 2000-89623 A Japanese Patent Laid-Open No. 5-323740 Japanese Patent Laid-Open No. 7-104619 JP 2004-219617 A JP 2001-175328 A

  For this reason, as disclosed in the above-mentioned patent document, a toner image formed on a photosensitive member or an intermediate transfer belt during normal image formation, or a toner image transferred onto a sheet and then read abnormally with an image sensor. Although there is a device to judge whether it is an image, an image evaluation device with a detection capability comparable to human vision needs to have extremely high calculation capability, and it is practical to use for such purposes Not right.

  Therefore, there is a device to quantify the status signal obtained by using the sensor output and the sensor or the operation control information, and to determine the normal state and the failure or the failure predictive state by a statistical mathematical method. However, such as Fischer linear type discrimination and cluster analysis In the method of creating a state index with the concept of distance from the state value, a method for determining the state value normalization method to properly create a homogeneous homogeneous space when creating a state index from the state value of different units with the concept of distance There is a limit to this, and satisfactory results cannot be obtained. In the MT method, the Mahalanobis distance is used and normalization is performed based on variations at normal times, but this idea may not be applicable.

  There is also a device that evaluates the sensor output and the status signal obtained using the sensor or operation control information qualitatively based on the knowledge base and diagnoses whether it is a pre-failure state or not. It is extremely difficult to construct a knowledge base that can properly evaluate the problem with a complex device such as an image forming device, and it takes a long time to detect a failure sign, which may result in useless operation during this time. However, it is still a subject of research, including the resolution of these new problems.

  On the other hand, the diagnosis result of the failure sign may not match the phenomenon of the device in the actual operation state. In this case, the diagnosis is performed by trial and error, so the diagnosis work becomes large and the diagnosis can be made substantially. New problems arise, such as the loss of control and the need for large-scale control devices for diagnosis, resulting in increased costs.

  Furthermore, depending on the installation location of the image forming apparatus, it is widely used in various environments such as various offices and factory environments, and the temperature / humidity, paper used, operation amount, image pattern, etc. are extremely diverse for each user. The device configuration process such as toner, developer, charger, photoconductor, and photoconductor cleaner used by the device is strongly affected by the various conditions described above, so there are cases where urgent maintenance due to failure is required. It just happens. For this reason, it is desirable to quickly grasp the failure status during emergency maintenance associated with the failure.

The object of the present invention has the following objects in view of the problem in the failure diagnosis in the conventional image forming apparatus.
The first purpose is to provide a method for determining the state of a practical device, in particular, a failure or a sign of failure, and knowing that the state has deteriorated before or immediately, thereby preventing unnecessary operation and systematically performing maintenance. It is an object of the present invention to provide an operating state determination method that can save time and resources required for failure recovery.

  The second objective is to realize a failure prediction method for operating state determination, and to create small classifiers and determine weights with sufficient accuracy based on sample data without relying on expert knowledge. It is an object of the present invention to provide an operating state determination method that can be obtained by a learning algorithm that can obtain the above.

  As a third object, there is provided an efficient operation state determination method capable of determining the state of an apparatus without a significant cost increase.

  A fourth object is to provide an image forming apparatus capable of accurately determining the state of the apparatus without a large cost increase and minimizing a loss associated with a failure.

  A fifth object is to provide a highly reliable electrophotographic image forming apparatus that does not require urgent maintenance due to a failure under any circumstances.

In order to achieve this object, the present invention has the following configuration.
(1) at least one sensor;
A state signal extractor for extracting a plurality of state signals from the sensor output;
Perform temporal feature extraction to calculate the temporal feature of the state signal,
Prepare a small classifier for each temporal feature,
The operating state is characterized in that the state of the operating device is determined by a weighted majority of the small discriminator discrimination results after performing a prior discrimination as a small discriminator using a stamp discriminator that compares and discriminates between a threshold value and a threshold value. How to determine.

(2) In the operating state determination method according to (1),
Use a learning algorithm based on sample data to create a small classifier and determine the weighting when voting.
For the small classifier, select the small classifier that minimizes the weighted misclassification rate each time the weight is repeatedly updated.
An operation state determination method characterized by determining the state of an apparatus by weighted majority of all selected small classifiers after a certain number of repetitive learning times.

( 3 ) An image forming apparatus using the operating state determination method according to any one of (1) and (2) .

( 4 ) In the image forming apparatus described in ( 3),
An image forming apparatus, wherein operation state determination is applied to a plurality of color image forming units.

According to the present invention, when determining a normal state and a failure or a predictive failure state by a statistical mathematical method, a small discriminator is previously provided for each state value instead of the concept of distance in a space formed by a state value group of different units. After pre-discrimination after considering the difference in physical meaning such as units, the state of the device is discriminated by weighted majority of the discrimination results, so the space created by the state value group of different units etc. It is possible to avoid problems such as quality and homogenization, and to handle it as an appropriate weighting factor determination problem when making a majority decision based on learning data. Become. Also, since temporal feature values are extracted and used for state determination, placing them in various environments and operating conditions results in extremely diverse states, but there are temporal changes in characteristics. Since it is determined that the state has occurred, it is possible to avoid erroneously determining a change in state due to use conditions as a failure or a failure sign state, and to perform highly reliable state determination.
This prevents wasteful operation that has been performed without noticing the failure so far, and eliminates downtime from when operation becomes impossible until completion of post-maintenance, and is necessary for failure recovery by performing planned maintenance. Time and resources can be saved.

  Further, according to the present invention, when performing learning repeatedly based on sample data, the misclassification rate is not repeatedly calculated uniformly for all data every time the weight is updated, but only on data that has failed to be discriminated. Since the calculation is performed using the weighted misclassification rate, the amount of calculation can be greatly reduced, and the problem that it takes enormous calculation time for learning and cannot be put into practical use can be solved.

  In addition, according to the present invention, since a so-called stamp discriminator that compares a feature amount with a threshold value is used for the small discriminator, the determination calculation of the small discriminator can be performed very simply, and the accuracy is determined with weighted majority. Since state determination is performed well, the state determination result of the device can be obtained in a short time even with a relatively small hardware resource such as a CPU or memory attached to the device, so that the cost effectiveness can be greatly improved and realized. .

  According to the present invention, since the apparatus state determination method according to claims 1 to 3 is installed in the apparatus, it is possible to quickly detect a failure of the apparatus and a predictive state of the apparatus with high accuracy without a significant increase in cost. The loss of repair and repair time can be minimized by preventing the generation of products and performing preparations and planned maintenance necessary for repairs.

  In addition, since the apparatus state determination method according to claims 1 to 3 is mounted on an electrophotographic image forming apparatus that forms an image by a complicated process such as charging, development, transfer, and cleaning, the above-described processes are optimally operated. Sensors and status values obtained from the sensors are internally used in advance, and can be used to accurately and accurately detect device failures and predictive failure states without significant cost increase. Based on the status judgment result, you can stop the operation immediately to prevent the occurrence of defective prints, or to minimize the loss of repair repair time, which has been a wide issue by preparing for the repair and planned maintenance, An image forming apparatus with extremely high reliability can be provided.

The best mode for carrying out the present invention will be described below.
FIG. 2 is a diagram showing a configuration of an image forming apparatus to which the operating state determination method according to the present invention is applied.
In FIG. 1, an image forming apparatus is a color printer having a plurality of color image forming units, and the outline thereof is as follows.

  The image forming apparatus 20 includes image forming apparatuses 21Y, 21M, 21C, and 21K that form images of the respective colors according to the original image, and a transfer disposed so as to face the image forming apparatuses 21M, 21C, 21Y, and 21K. The apparatus 22, the image forming devices 21M, 21C, 21Y, 21BK, and the transfer device 22 face each other, and a paper feeding device 24 that feeds the recording paper from the paper feeding cassette 23 toward the transfer area. In the transfer region, a registration roller 30 that feeds the registration timing set in accordance with the image formation timing by the devices 21M, 21C, 21M, and 21K, a secondary transfer device 40 that collectively transfers the superimposed image transferred to the transfer device 22, and a transfer region. Are provided with a fixing device 60 and a paper discharge device 70 for fixing the image on the recording paper on which the image is transferred. In the following description, black may be referred to as black (Bk).

  The image forming apparatuses have the same configuration. In FIG. 2, the image forming apparatus 21K for black images will be described. The photosensitive drum 21K1, the charging apparatus 21K2, the developing apparatus 21K3, and the cleaning apparatus 21K4 are combined into a process cartridge. And the photosensitive drum 21K1 is disposed to face the transfer belt 22A included in the transfer device 22. In FIG. 2, reference numeral 80 denotes a writing device used in an exposure process for each image forming device.

The transfer device 22 is a primary transfer unit that sequentially transfers the color images formed in each image forming device. The image superimposed and transferred to the transfer belt 22A is conveyed by the registration roller 30 with the registration timing set. The incoming recording paper is batch-transferred by the secondary transfer device 40 corresponding to the secondary transfer unit and conveyed toward the fixing process.
After the batch transfer, the transfer belt 22A is removed of foreign matters such as residual toner and paper dust by a transfer device cleaning device.

A patch pattern image P1 (see FIG. 3) is formed on the transfer belt 22A used in the transfer device 22 for image density detection and transfer position deviation detection between images.
FIG. 3A is a diagram showing the configuration of the transfer device 22. The transfer belt 22 </ b> A having a stretched surface between the image forming devices has five levels for each color at two positions along the stretch direction. A toner image corresponding to the patch pattern image P1 having the toner density of is formed. In FIG. 3B, for convenience, a toner image of five levels (dark and light) and a density difference in hatched density is shown for two types of colors. A five-step pattern for four colors is formed.

This toner image is detected by a toner image density sensor 50 (see FIG. 2) disposed in the vicinity of the transfer belt 22A.
As shown in FIG. 5A, the toner image density sensor 50 is a reflective optical sensor that detects the amount of light reflected from the toner image, and includes an LED light source 50A, a regular reflection photodiode 50B, a light source LEFD 50A, and a regular reflection photodiode. The diffused reflection photodiode 50C disposed between the two and 50B is combined. In the following description, the photodiode is indicated as PD.

  FIG. 4 is a block diagram illustrating a configuration of the control unit 100 in which the toner image density sensor 50 is connected to the input side. In FIG. 4, the toner image density sensor corresponds to the case where the toner image is formed at two locations for the case shown in FIG. 3, and the regular reflection photodiode 50 </ b> B and the irregular reflection photodiode 50 </ b> C are provided. Although only two are shown, it is of course that they are actually matched to the number of image forming devices.

  In the figure, when the normal operation signal is instructed by the host controller of the image forming apparatus, the control unit 100 (indicated by CPU in the figure) starts the image signal generating circuit and the exposure laser diode blinks in response to the image signal. To do. In addition, the CPU sequentially outputs a bias means for each image forming process including a driving means such as a photosensitive motor and a charging bias, and executes image formation. In the control unit 100, in the case of an electrophotographic image forming apparatus, there is a weak point that the image density fluctuates due to deterioration with time or environmental fluctuations. Therefore, a process control sensor such as a toner image density sensor is provided to stabilize the image density. Control.

  The control unit 100 executes a process control operation when the image forming apparatus is started up or when a predetermined number of prints are performed, that is, a process control operation for updating an initial state in an apparatus used for so-called image formation. When the process adjustment operation signal is instructed by the host control, the process control operation is started when the normal operation signal is received or the time point after the image forming operation is performed by the normal operation signal. It has become.

In the process control operation, the calibration process of the toner image density sensor 50 is first performed. In other words, the image signal generation circuit shown in FIG. 4 is in an image-free state, and the control unit (CPU) 100 adjusts the amount of emitted light so that the light reception signal from the surface of the photosensitive drum becomes a predetermined value.
As a result, the toner image density can be accurately measured without being affected by variations in light receiving and emitting elements, changes over time, and changes in the surface state of the photoreceptor.

  Next, a specific test image using the toner image forming the patch pattern shown in FIG. 3 is automatically output, and the toner image density sensor 50 optically adjusts the density and position of the corresponding toner image on the photoreceptor. measure.

  FIG. 7 is a flowchart for explaining the contents of the above-described process control operation. The amount of light reflected by the toner image density sensor 50 is measured in a state where there is no image on the transfer belt 22A, and the light source LED is set so that the predetermined reflection amount is obtained. Adjust the lighting status of the. When the adjustment is completed, a toner image having the patch pattern shown in FIG. 3 is formed and optically measured by the toner image density sensor 50. As described in FIG. 3, the toner image used as the test image used for the patch pattern image is a pattern having a uniform density that has been exposed at about five levels with different density levels. At this time, the charging and developing bias conditions are executed with predetermined specific values.

  As shown in FIG. 8, when a development potential-toner adhesion amount straight line that is linearly approximated from five received light signals is obtained, the slope and intercept may deviate from the target characteristics due to fluctuation factors such as deterioration with time and environmental fluctuation. Since it can be detected, the inclination is mainly corrected by multiplying the exposure light amount correction parameter P by the exposure signal, and the deviation of the potential (intercept X0) at which development is started is mainly by multiplying the development bias by the correction parameter Q. It is possible to stably obtain the target image density.

  As shown in FIG. 5, the toner image density sensor 50 includes a light source LED 50A, a regular reflection PD 50B, and an irregular reflection PD 50C. In order to avoid sticking, smoothness on the surface is important. For this reason, as the belt material, a material having a glossy surface such as PVDF or polyamide is selected.

In the toner image density sensor 0 intended for the transfer belt 22A using the above materials, as shown in FIG. 5A, the luminous efficiency individual difference of the LED using the regular reflection PD50B installed on the regular reflection optical axis. As shown in FIG. 6B, the control current to the light source LED is controlled so that a constant amount of received light can be secured as shown in FIG. 6B.
Thereafter, when light scattered by the toner image is received by the irregular reflection PD 50 </ b> C that mainly receives irregular reflection light when the toner image comes to the opposite position, the toner changes from the change shown in FIG. Measure the concentration. Since the toner contains a colorant of each color, the light source LED 50A uses a near-infrared or infrared light source having a wavelength of about 840 nm that is not significantly affected by the colorant. However, as black toner, toner colored with low-cost carbon black is generally used and exhibits strong absorption even in the infrared region. Therefore, as shown in FIG. 6B, sensitivity to toner density is lower than other colors. Become.

  In this example, the exposure light amount and the development bias are corrected. Of course, other process control values that contribute to the image density such as the charging potential and the transfer current may be corrected to obtain the same result.

These process controls are operated for the purpose of correcting fluctuations in the toner charge amount within the normal range due to temperature and humidity and fluctuations in the sensitivity of the photoconductor, but they are also measured when there is a specific failure or a sign of failure. The parameter determined based on the value or the measured value may vary. For example, the cleaner installed for the purpose of collecting the toner remaining without being transferred onto the photoconductor after the transfer and maintaining normal charging exposure is often used with a blade cleaning system that rubs the photoconductor with a urethane rubber blade. However, because of such a configuration, a part of the toner sinks under the blade and passes.
The toner that has passed passes through the charged exposure part and is recovered by development, but the charging characteristics are lost due to frictional action by the blade or the shape is changed, so that the toner is not recovered by development and the image is transferred to the transfer body. Regardless of whether it is a part or a non-image part, there are some which adhere non-electrostatically and are transferred as they are.

For this reason, as shown in FIG. 9A, a very small amount of toner particles are also adhered to the non-image area, but the image quality is not seriously deteriorated because the amount is very small.
When the photosensitive member contact portion of the blade is worn due to long-term rubbing, the scraping force decreases, and the amount of passing toner tends to increase at an accelerated rate. Finally, when a large amount of residual toner passes through the blade at once, the charging device deteriorates the charging ability due to contamination by the toner, the function of the exposure unit also deteriorates due to the attenuation by the toner, and the developing unit also has such a large amount of toner. Can not be recovered, and finally a very unacceptable vertical-shaped abnormal image is generated, which immediately requires repair.

  By the way, from a point in time just before reaching such a state, the toner adhesion amount has increased substantially uniformly over the entire image carrier as shown in FIG. 9B. There is no noticeable image degradation and there is very little to notice. This state is called “light soiling” and is considered to be a sign of cleaner failure.

  As shown in FIG. 10, the presence of such toner has an effect of increasing the measurement result particularly in the low density portion and causes a slight decrease in the slope γ and a decrease in the intercept X0. There is no significant difference from the environmental aging fluctuation range, and it is very difficult to discriminate this occurrence from fluctuations of monochromatic γ and X0 or fluctuations of correction parameters P and Q determined based on this, and it is difficult to make a predictive alarm with high accuracy. It is.

  Because of this difficulty, the conventional apparatus clearly only gives a failure alarm when it deviates from normality, and it is difficult to quickly find an abnormality at the level of failure prediction.

With reference to the calibration as described above, an embodiment of the invention described in claim 1 will be described with reference to FIGS.
FIG. 1 shows a system configuration used when carrying out the operating state determination method according to the present invention. Note that the reference numerals shown in FIG. 1 are different from the reference numerals used in the previous drawings for convenience.
In FIG. 1, an image forming apparatus 901 has sensing signals such as the correction parameters P and Q described above. That is, the control device 100 and its operation shown in FIG. 4 are used as a state signal extractor, and the correction parameters P, Q and the like correspond to the state signal. This is recorded by the data collector 902 in the form of a log in the step indicated by reference numeral 903, and whether the temporal feature quantity extractor 904 shows a singular change with respect to the past signal movement is mathematically or statistically determined. To calculate the condition data set at that time.

  For example, when the density characteristics of each color are obtained as shown in FIG. 11, the log of the correction value Q is updated as shown in FIG. The approximate differential value dQ can be obtained by dividing the difference of the Q value at the previous time by the elapsed time or the elapsed operation amount.

Since it is considered that the deterioration of the apparatus is mainly controlled by the operation amount, it is preferable to employ the operation time or the printed sheet counter value as the elapsed operation amount. Since these operation amounts are generally internally managed by the CPU, the data collector can record the operation amounts together.
The data collector may be configured using the control unit (CPU) 100 shown in FIG. 4 and memory means (not shown) configured accompanying the control unit (CPU) 100 or connected to the control unit (CPU) 100 by communication. You may implement | achieve with another control part (CPU) and a memory means. For example, a log is collected via a communication means such as a network on a management apparatus of an image forming apparatus generally called a controller that controls the image forming apparatus at a higher level, or a dedicated monitoring server provided separately from the apparatus and the controller, It may be recorded.

  In addition to such differential values as described above, various temporal feature values such as a regression value of signal change, a standard deviation, a maximum value, and an average value of a plurality of recent data can be obtained. Many such time-series signal feature extraction methods such as the ARIMA model have been proposed, and an appropriate method may be used. In addition, the time lapse index is not limited to the print sheet counter value as described above, and an operation time integrated value, an actual time elapsed value, or the like can be selected.

  The sign of failure is considered to be captured by a signal that was stable when normal but in various forms but showed unusually unstable movements. From this perspective, an appropriate temporal feature extraction method should be used. Just choose.

In addition, adding the feature quantity not including time calculation to the condition data set does not impair the advantage obtained by the present invention.
For example, the sensing signal value itself at that time may be added. Or you may add driving | operation information, such as the driving | running time used for judgment of a lifetime, or a real time elapsed value.

  A signal indicating that failure repair has been performed may be prepared and added to the log, and exception processing may be performed so that a transient change in the condition data set immediately after repair is not erroneously determined as a failure sign state.

The condition data set obtained by the above procedure is sent to the discriminator 906.
Discriminator, corresponds to step ST 4 procedure after (algorithm) shown in FIG. 12, the condition data set to determine whether normal or failure prediction state. That is, a normal discriminator is discriminated by a small discriminator provided for each condition data (temporal feature quantity such as dQ), and then a discriminant index value F is calculated by a weighted majority rule. It is determined that the state is a prediction state, and a user or a service station is notified through the alarm notification device 907.

  The stamp discriminator has the merit that the CPU operation can be performed at a very high speed, and this method using the weighted majority vote provides a sufficient accuracy, so that it is very suitable for realizing the failure prediction technology with high accuracy and without cost. It is a small classifier. This is the effect of the invention according to claim 3.

The state discrimination calculation method when a stamp discriminator is used as the small discriminator is as follows.
A stamp discriminator is prepared for each of the n temporal feature value calculation results C1 to Cn for P, Q, and R, and a weighted majority calculation result F value is obtained by the following equation (1).

The small stamp discriminator is as follows.
Outi = 1 (sgn i × (Ci−bi) ≧ 0)
Outi = −1 (sgn i × (C i −bi) <0)
(Where bi is the threshold value for each feature, and sgn i is the discrimination polarity)
When the F value obtained in this way is smaller than 0, it may be determined as a failure sign state.

In the invention according to claim 2, in order to determine the discriminant condition creation (determination of bi and sgn i) of the small discriminator used in such a failure prediction method, the calculation method of weighted majority (determination of αi) is generally performed in a booth. A supervised learning algorithm called “ting method” may be used. For a description of the boosting method, see Mathematical Sciences 489, MARCH 2004 “Statistical Pattern Identification Information Geometry”.
That is, first, sensing log data that is known in advance to be in a normal state and sensing log data that is known to be in a failure sign state are prepared. For example, a sensing data log is taken when performing an endurance test of the device, and when a failure example is encountered, a period of a predictive state before the failure is estimated and used as the data.

FIG. 14 and subsequent figures show the results of verification by collecting failure cases while actually taking the sensing data log over three months by the present inventors for more than 10 image forming apparatuses.
FIG. This is the result of recording the change in the four-color parameter Q (the value corresponding to the intercept Xo with the opposite sign) in the case of a failure case in which the 833 machine is repaired with Bk color causing a defective cleaning.

Similarly, a number of other parameter changes are recorded and utilized, but only the parameter Q for which the change is remarkable will be introduced. In this way, it is observed that the parameters Q of the Y, M, and C colors fluctuate prior to the poor cleaning of the Bk color.
The above temporal feature extraction was performed to extract these changes and generate a condition data set.

  Visually estimate the failure sign period, give -1 (failure sign period) for the relevant part of the condition data set, and 1 (normal period) for the other part of the condition data set. b1 to b100, sgn1 to sgn100, and α1 to α100 were determined. FIG. 15 shows the result of calculating the F value using the data used for learning.

  It was confirmed that the supervised data with the label was appropriately learned, and a small discriminator in which only the portion corresponding to the sign changed to a negative F value and a discriminator by weighted majority were generated.

  Next, create a condition data set in the same procedure from the sensing log data of the device where five similar failure cases occurred to determine whether this discriminator gives an appropriate result to the test data not used for learning. The verification result is shown in FIG. In FIG. 16, the machine number boiled on the left side of the diagram means the number of the testing machine.

The discriminator output F value calculated by b, sgn, and α determined in advance is changed to minus when the failure predictor is in a state prior to the occurrence of the same failure case as intended, and the predictor is successfully We were able to confirm that we were able to distinguish.
In other words, if an actual device is operated while operating such a state discriminating method using a temporal feature quantity extractor or discriminator, it is possible to know that the state discrimination result is in a failure sign state through the F value. In other words, it was shown that it can be avoided by exchanging and repairing the failed unit before the occurrence of vertical abnormal images.

  Such a change in state before the failure is considered to be a phenomenon that can occur universally not only in an image forming apparatus but also in an apparatus having an autonomous control means equipped with all sensors. If it is installed, it becomes clear that loss due to failure can be avoided in advance, and even when using the actual machine, it is possible to make highly accurate judgments such as predictive state, so that downtime due to sudden failure occurrence is minimized It was recognized that there was an effect in reducing it.

1 is a system configuration diagram for implementing an operating state determination method according to the present invention. FIG. 2 is a diagram illustrating an image forming apparatus to which the operation state determination method illustrated in FIG. 1 is applied. It is a figure for demonstrating the structure of the transfer apparatus used for the image forming apparatus shown in FIG. It is a block diagram for demonstrating the structure of the control part used for the operating state determination method by this invention. It is a figure for demonstrating the reflection state by the structure of the sensor used for the transfer apparatus shown in FIG. 3, and the adhesion state of a toner. FIG. 6 is a diagram for explaining a detection result regarding a light amount by the sensor shown in FIG. 5 and a density obtained from the detection result. It is a flowchart for demonstrating the process control operation performed in the control part shown in FIG. FIG. 8 is a diagram for explaining a method of adjusting various image forming conditions executed during the process control operation shown in FIG. 7. FIG. 6 is a diagram for explaining a failure sign corresponding to the amount of toner attached on the image carrier. It is a diagram for explaining the degree of contamination on the image carrier due to environmental fluctuations and the adjustment method. It is a well line figure for demonstrating the fluctuation | variation state of each color in the dirt condition shown in FIG. It is a flowchart for demonstrating the procedure implemented in the system configuration | structure shown in FIG. It is a diagram for demonstrating the majority decision performed within the procedure shown in FIG. FIG. 2 is a diagram showing changes in parameters for each color for a failure example caused by a cleaning failure in the system configuration shown in FIG. 1. It is a diagram which shows the data used for the learning at the time of performing the data update for failure sign discrimination | determination. FIG. 4 is a diagram showing a failure due to a signal from the sensor shown in FIG. 3 and data after failure repair.

Explanation of symbols

20,901 Image forming apparatus 22A Transfer belt 50 Toner image density sensor 51 LED
100 Control Unit P1 Toner Image Forming Patch Pattern 902 Data Collection Machine 904 Extractor 906 Small Discriminator

Claims (4)

At least one sensor;
A state signal extractor for extracting a plurality of state signals from the sensor output;
Perform temporal feature extraction to calculate the temporal feature of the state signal,
Prepare a small classifier for each temporal feature,
The operating state is characterized in that the state of the operating device is determined by a weighted majority of the small discriminator discrimination results after performing a prior discrimination as a small discriminator using a stamp discriminator that compares and discriminates between a threshold value and a threshold value. How to determine. The operation state determination method according to claim 1,
Use a learning algorithm based on sample data to create a small classifier and determine the weighting when voting.
For each small classifier, select a small classifier that minimizes the misclassification rate each time the weight is repeatedly updated.
An operation state determination method characterized by determining the state of an apparatus by weighted majority of all selected small classifiers after a certain number of repetitive learning times. An image forming apparatus using the operation state determination method according to claim 1. The image forming apparatus according to claim 3 .
An image forming apparatus, wherein operation state determination is applied to a plurality of color image forming units.

Images