JPH0934188A - Image forming device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To stabilize the toner density in a developing unit, and to stabilize the image density, by processing various factor fluctuation concerning the image density in electrophotographic process altogether. SOLUTION: The toner sticking amount to a test toner image formed on a photoreceptor drum 21 under the specific image forming condition is detected by an AIDC sensor 51, and the toner is fed to the developing unit 23 based on the detected value. At the same time, the toner density is detected by the ATDC sensor 54 provided in the developing unit 23. The detected toner density is corrected based on information such as humidity and the number of copies, applying fuzzy inference or a neural network. Based on the corrected toner density, the electrophotographic process (electrification potential, developing bias voltage and image exposing quantity) is respectively controlled.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image forming apparatus, and more particularly to an image forming apparatus such as an electrophotographic copying machine, a laser printer or a facsimile which transfers a toner image formed on an image carrier by an electrophotographic process onto a sheet. .

[0002]

2. Description of the Related Art In an image forming apparatus using an electrophotographic process, stable formation of an image on a photosensitive member is an important issue, and toner concentration in a developing device (toner relative to the total weight of carrier and toner) It is necessary to maintain the electrophotographic process such as the charging potential of the photoconductor, the developing bias voltage or the image exposure amount to an appropriate value.

In the conventional toner replenishment control system, an optical sensor detects the toner adhesion amount based on the amount of reflected light from a test toner image formed on a photoconductor under a constant image forming condition. A so-called AIDC that estimates the toner concentration in the developer from the adhered amount and replenishes the required amount of toner
And a so-called ATDC that supplies magnetic toner in a required amount by estimating the toner concentration in the developer by detecting the magnetic permeability of the developer by a magnetic sensor or detecting the amount of reflected light from the developer by an optical sensor. Are known.

[0004]

By the way, the AID
In C and ATDC, influences such as changes in environmental conditions such as humidity and the number of copies (endurance of carrier) cannot be ignored. From a past experience and an experiment, it has been found that there is a certain correlation between the image density and the factor affecting it. Therefore, if the detected value by AIDC and / or ATDC is corrected based on the variation of various factors, the image stabilization control can be performed well. However, the correlation between the factor variation and the image density is complicated, and the correction process becomes complicated.

Therefore, it is an object of the present invention to collectively process the fluctuations of various factors relating to the image density in the electrophotographic process to stabilize the toner density in the developing device and the image density. An object is to provide an image forming apparatus.

[0006]

In order to achieve the above objects, the image forming apparatus according to the present invention first comprises a toner of a test toner image formed on an image carrier under a constant image forming condition. A first sensor for detecting the amount of adhesion and a toner replenishing means for replenishing the toner to the developing device based on the detection value of the first sensor are provided, and the toner is replenished by AIDC.
Further, a second sensor that detects the toner concentration in the developing device, a third sensor that detects a variation factor of the detection value of the second sensor (for example, humidity, history of use of the developing device), and a second sensor
And a toner density correction means for determining the toner density correction amount based on the detection value of the third sensor, and a control means for controlling the electrophotographic process based on the toner density corrected by the toner density correction means. ing.

That is, in the image forming apparatus according to the present invention, toner is supplied to the developing device by AIDC so as to keep the image density constant. In this case, the toner concentration in the developing device fluctuates due to factors such as humidity and the history of use of the developing device (carrier durability), and the toner concentration may exceed the allowable range. Therefore, the ATDC is also used, that is, the toner density in the developing device is detected by the second sensor, and the electrophotographic process (charging potential, developing bias voltage, or image exposure amount) is controlled to stabilize the image density. At the same time, the toner density in the developing device is maintained within a certain range.

Since the detection value of the second sensor depends on the fluctuation of the factor, the detection value of the second sensor is corrected by the detection value of the third sensor which detects the fluctuation of the factor. This correction means uses fuzzy inference or a neural network.

According to the present invention, the toner is supplied by the AIDC and the electrophotographic process is controlled by the ATDC, so that the toner density in the developing device is maintained within a certain range and the image density is stabilized. be able to. Moreover, since the detection result by ATDC is corrected based on the detection value of the third sensor, not only the true toner concentration can be obtained, but also the correction is performed by using fuzzy inference or a neural network. It is possible to apply the knowledge and know-how obtained in step 1, and collectively process the factor fluctuations to accurately determine the true toner concentration.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of an image forming apparatus according to the present invention will be described below with reference to the accompanying drawings.

(Overall Structure of Copying Machine and Outline of Copying Operation) FIG. 1 shows a digital copying machine which is an embodiment of the present invention. In this copying machine, the image reader unit 1 is arranged in the upper stage, the laser scanning unit 10 is provided immediately below the image reader unit 1, the image forming unit 20 is arranged in the middle stage, and the sheet conveying system 40 is arranged in the lower stage.

The image reader unit 1 moves in the direction of arrow a to read a document image set on the platen glass 9 by a scanner 2, and image signal processing for converting the read image data into printing data. And the part 5. The scanner 2 is a well-known one having a contact type line image sensor (CCD) 3.

In the laser scanning unit 10, the light source driving unit 11 modulates and drives the light source (laser diode) based on the print data generated and edited by the image signal processing unit 5,
An electrostatic latent image is formed on the photosensitive drum 21. The laser scanning unit 10 includes a polygon mirror 12 and an fθ lens 1
It is composed of 3 etc., and its structure and action are well known.

The image forming unit 20 is constructed around a photosensitive drum 21 which rotates in the direction of an arrow b, around which a charging charger 22, a developing device 23, a transfer charger 31,
Sheet separation charger 32, residual toner cleaner 3
3. An eraser lamp 34 for residual charge is arranged.
Further, an AIDC sensor 51 for optically detecting the toner adhesion amount of the test toner image, a humidity sensor 52 for detecting the humidity inside the machine, and a temperature sensor 53 for detecting the temperature are arranged.

The developing device 23 uses a two-component developer composed of a mixture of carrier and toner, and the toner is contained in a hopper 24, and is supplied to the stirring / circulating tank 26 by the rotation of a supply roller 25. To be done. The replenished toner circulates in the stirring / circulation tank 26 while being stirred, and is charged to a predetermined electric charge by friction with the carrier, and is supplied to the surface of the photosensitive drum 21 through the developing sleeve 27. Further, an ATDC sensor 54 for magnetically or optically detecting the toner concentration in the developer is arranged in the stirring / circulating tank 26.

The sheet conveying system 40 feeds the sheets accommodated in the sheet feeding cassette 41 one by one by the rotation of the sheet feeding roller 42, and the toner image formed on the photosensitive drum 21 by the timing roller 43. It is sent to the transfer section in synchronization. At the transfer section, the sheet is the transfer charger 3
The toner image is transferred by the discharge from No. 1 and is separated from the photoconductor drum 21 by the charge removal by the AC discharge from the sheet separation charger 32. The separated sheet is sent to the fixing device 47 by the air suction belt 46 to fix the toner image, and is discharged from the discharge roller 48 onto the tray 49.

(Control Mechanism) FIG. 2 shows a control mechanism of the copying machine. The control mechanism mainly includes a CPU 60 and a toner density correction unit 70. The CPU 60 is a control ROM that stores a control program, a data ROM that stores various data, and a RAM that stores various parameters.
It has. The data ROM stores various data necessary for performing toner density control and image density control described later. The detection value from the AIDC sensor 51 is input to the CPU 60. Further, the CPU 60 is the light source drive unit 1.
1, the grid voltage transformer 71, the developing bias transformer 72, the developing device 23, and the toner hopper 24 are controlled.

On the other hand, the toner concentration correction unit 70 has a durability counter 55 for integrating the detection values of the humidity sensor 52, the temperature sensor 53 and the ATDC sensor 54, and the driving time of the developing device 23.
The count value from is input.

(Stabilization of Image Density and Replenishment of Toner) In the copying machine described above, the photosensitive drum 21 is charged by
This is performed by applying the grid voltage Vg from the transformer 71 to the grid 22a of the charging charger 22 having the discharge voltage Vc. Charge potential Vo of the photosensitive drum 21 before exposure
Is approximately equal to the grid voltage Vg, and the charging potential Vo can be controlled by changing the grid voltage Vg.

In this embodiment, the laser scanning unit 10 is used.
Is exposed to a laser beam emitted from a low potential Vi
The so-called reversal development is used in which the toner is attached to the image portion where the voltage becomes almost 0V. If the charging polarity of the photoconductor is negative, the charging polarity of the toner is also negative, and the developing bias voltage Vb of the negative polarity is applied to the developing sleeve 27 of the developing device 23 from the transformer 72. In the reversal development, the toner adheres to a potential portion lower than the development bias voltage Vb. The development potential difference is the difference between the development bias voltage Vb and the image portion potential Vi. In the reversal development, if the image potential difference is set to be large, the development efficiency will be high, and if it is set to be small, the development efficiency will be low. here,
The developing efficiency refers to the amount of toner adhering to the photosensitive member per unit development potential difference.

In this embodiment, after the copying operation for one sheet is completed, a test toner image is formed on the photosensitive drum 21 under a predetermined grid voltage Vg, developing bias voltage Vb and exposure amount, and the AIDC sensor 51 tests the test toner image. The scattered reflected light of the image is detected. This output voltage is input to the CPU 60 and the toner adhesion amount is calculated. AIDC sensor 5
An example of the relationship between the output voltage of No. 1 and the toner adhesion amount of the test toner image is as shown in FIG.
If it is less than 8 mg / cm ² , the image density will decrease,
The CPU 60 drives the hopper 24 to apply the toner to the developing device 23.
Supply to.

By the way, when the toner is replenished by the AIDC to prevent the image density from being lowered, the toner density in the developing device 23 may vary greatly. For example, when the toner density rises more than necessary, (1) when the high humidity state is reached, the image portion potential Vi on the photosensitive drum 21 rises, the developing potential difference decreases, and the developing efficiency decreases, so that the toner Excessive supply. (2) When the humidity is low, the toner charge amount increases and the developing efficiency decreases, so that the toner replenishment becomes excessive. On the other hand, when the toner concentration is reduced more than necessary, (1) if the humidity rises and the toner charge amount decreases, the developing efficiency is too high, so that the toner replenishment is suppressed. (2) When the toner charge amount decreases due to carrier deterioration due to an increase in the number of copies, development efficiency increases, and toner supply is suppressed.

A particular problem here is when the toner density rises abnormally. In this case, the amount of toner that is not properly charged is increased, the toner is fogged on the image background portion, and the inside of the machine is contaminated due to toner scattering. Therefore, in the present embodiment, the toner density is detected by the ATDC sensor 54,
If the toner density exceeds a certain value, the grid voltage Vg,
At least one of the developing bias voltage Vb and the exposure amount is controlled to set the developing potential difference to a large value. As a result, the development efficiency is increased, toner consumption is promoted, and excess toner in the developing device 23 is eliminated.

However, the toner density detection value by the ATDC sensor 54 varies depending on factors such as environmental conditions and the number of copies. It is considered that this is because the fluidity of the developer changes depending on the humidity and the durability of the carrier, and the amount of the developer accumulated on the detection surface of the ATDC sensor 54 changes. FIG. 4 shows the relationship between the output voltage of the ATDC sensor 54 and the toner concentration when the absolute humidity is 2 g / m ³ , 10 g / m ³ , 20 g / m ^3.
It shows about when. FIG. 5 shows the relationship between the output voltage of the ATDC sensor 54 and the number of copies.

Here, the relative humidity and the absolute humidity will be described. Amount of water vapor actually contained in a certain volume of air e
And the ratio of the saturated vapor amount E of the air to the ratio [(e / E) × 100] is the relative humidity. On the other hand, the absolute humidity is the amount of water vapor contained in the air having a volume of 1 cubic meter expressed in g / m ³ . The absolute humidity is calculated from the temperature, saturated water pressure above that temperature, and relative humidity.

In the present embodiment, the saturated water vapor pressure is obtained from the detected values of the humidity sensor 52 and the temperature sensor 53 by referring to the data table stored in the data ROM, and the absolute humidity is obtained from the following calculation formula. A = (0.01058 × H × P) / (1 + 0.0036
6 × T) A: Absolute humidity (g / m ³ ) H: Relative humidity (%) T: Temperature (° C.) P: Saturated water vapor pressure at temperature T (mmHg)

In the present embodiment, the ATDC sensor 5
The toner concentration predicted from the output voltage of 4 is measured by the sensor 5
The correction is made by the absolute humidity determined based on the detected values of 2, 53 and the number of copies converted from the count value of the durability counter 55. Based on the corrected correct toner density, at least one of the grid voltage Vg, the developing bias voltage Vb, and the exposure amount is controlled to switch the developing potential difference, thereby preventing excessive replenishment of toner by the AIDC.

(Control Procedure) Next, a control procedure for controlling the toner density will be described with reference to FIGS. 6 and 7. In FIG. 6, first, in step S1, a copy operation for one image is performed, and then in step S2, the photosensitive drum 2
A test toner image is formed outside the image forming area on 1 at a predetermined grid voltage Vg, exposure amount and developing bias voltage Vb.

Next, in step S3, the AIDC sensor 51
The toner adhesion amount of the test toner image is detected on the basis of the output voltage from, and necessary toner replenishment is performed on the basis of the detected value. That is, if the detected value is smaller than the reference toner adhesion amount, the toner is replenished, and if the detected value is large, the toner is not replenished. Next, in step S5, the toner density in the developing device 23 is detected. The toner density detected here is different from the toner density detected by the ATDC sensor 54 by the sensor 5
The toner density is corrected by the absolute humidity obtained based on the detected values of 2, 53 and the number of copies converted from the count value of the durability counter 55. Next, step S
At 6, it is determined whether or not the development potential difference is switched based on the corrected toner density. That is, when the corrected toner density is out of the allowable range, stabilization of the image density and excess toner,
The development potential difference is controlled in order to prevent the difference.

FIG. 7 shows a subroutine for detecting the corrected toner density, which is executed in step S5. Here, in step S11, the toner density detection value of the ATDC sensor 54 is fetched into the toner density correction unit 70, and in step S12, the toner density detection value is toner density Tc1 at an absolute humidity of 10 g / m ³ at the initial stage of durability (copy number 0). Convert to. Next, in step S13, the absolute humidity is calculated based on the detection values of the sensors 52 and 53. Step S14
In step S15, the count value of the durability counter 55 is fetched, and in step S15 the count value is converted into the number of copies.

Next, at step S16, the toner concentration Tc1
Then, the toner density correction amount Tc2 is determined from the absolute humidity and the number of copies by using fuzzy reasoning. Next, step S
In step 17, the toner concentration Tc1 and the correction amount Tc2 are added to calculate the true toner concentration Tc. In step S18, the toner concentration Tc is output from the toner concentration correction unit 70 to the CPU 60.

(Fuzzy Inference) Here, the step S
The fuzzy inference performed in 16 will be described. In the fuzzy reasoning here, the toner density correction amount Tc2 is obtained from the toner density Tc1 detected by the ATDC sensor 54, the absolute humidity, and the number of copies. The main control rules are
From the relationship between the toner concentration at each absolute humidity shown in FIG. 4 and the output voltage of the ATDC sensor 54, and the relationship between the number of copies and the output voltage of the ATDC sensor 54 shown in FIG.

(1) The correction amount increases to the negative side as the absolute humidity decreases. (2) The correction amount increases to the positive side as the absolute humidity increases. (3) The absolute value of the correction amount is larger when the toner density is higher than when it is low. (4) When the number of copies increases, the correction amount increases to the negative side. In addition to (1) to (4) described above, control rules are prepared based on various experimental data and experience of the experimenter, and fuzzy reasoning is constructed.

The state quantity as the input and the control quantity as the output of the fuzzy inference are as follows. Input (state amount): number of copies toner concentration Tc1 absolute humidity output (control amount): toner concentration correction amount Tc2 As a membership function, as shown in FIGS. 8 to 11, the fuzzy set of the state amount and the control amount is used. Define as a membership function. The symbols shown in FIGS. 8 to 11 are respectively ZO: standard PS: a little higher PL: very large in the number of copies in FIG. 8, NL in the toner concentration Tc (initial durability / standard humidity) of FIG. : Low ZO: Standard PL: High At absolute humidity in FIG. 10, NL: Low ZO: Standard PL: High At toner concentration correction amount Tc2 in FIG. 11, NL: Very low NM: Low NS: Slightly low ZO: Standard PS : Somewhat high PM: High PL: Very high

Further, in each of FIGS. 8 to 11, the vertical axis of the graph represents the certainty factor of the fuzzy set of each symbol,
It takes an arbitrary value in the range of 0 to 1. For example, when the number of copies is 12,000 as shown in FIG. 8, ZO and PS are selected as state quantities, and the ZO certainty factor is 0.2.
Then, the certainty factor of PS becomes 0.8. Further, as shown in FIG. 9, when the toner concentration Tc1 is 6%, ZO and PL are selected as state quantities, the ZO certainty factor is 0.8, and P is P.
The certainty factor of L is 0.2. Further, as shown in FIG. 10, when the absolute humidity is 8 g / m ³ , NL and ZO are selected as the state quantities, the NL certainty factor is 0.2 and the ZO certainty factor is 0.8. . In this way, the certainty factor of each state can be obtained for a certain input value from the membership function.

The control rule in the fuzzy inference is that the number of copies: Z, as shown in Tables 1a, 1b and 1c below.
In O, PS and PL, it is expressed in a matrix form with respect to humidity and toner concentration. The total number of rules is 27, and the control state quantity is determined with respect to the input state quantity.

[0037]

[Table 1]

For example, as described above, the number of copies is 1.
ZO and PS are selected as the state quantities for 2,000 sheets, ZO and PL are selected as the state quantities for a toner concentration of 6%, and the state quantities for humidity of 8 g / m ³ are selected. If NL and ZO are selected, Tables 1a, 1b, 1c
The control rules applied from the rule matrix of are the eight rules shown in Table 2 below.

[0039]

[Table 2]

From the control rules selected as described above, the control amount is min-max based on the membership function of the control amount.
Calculated by the center of gravity method. The certainty factor of the control amount in each selected rule is determined by the rule: certainty factor of copy number ZO = 0.2 certainty factor of toner concentration ZO = 0.8 certainty factor of humidity NL = 0.2 The assertion is that the certainty factor of the toner density correction amount NM is 0.2.

Rule: Confidence of copy count ZO = 0.2 Confidence of toner density ZO = 0.8 Confidence of humidity ZO = 0.8 Therefore, the assertion of the rule is confidence of toner density correction amount ZO = 0. .2.

Rule: Confidence of copy count ZO = 0.2 Confidence of toner density PL = 0.2 Confidence of humidity NL = 0.2 Therefore, the assertion of the rule is the confidence of toner density correction amount NL = 0. .2.

Rule: Confidence of copy count ZO = 0.2 Confidence of toner density PL = 0.2 Confidence of humidity ZO = 0.8 Therefore, the assertion of the rule is the confidence of toner density correction amount ZO = 0. .2.

Rule: Confidence of the number of copies PS = 0.8 Confidence of toner density ZO = 0.8 Confidence of humidity NL = 0.2 Therefore, the assertion of the rule is that the confidence of the toner density correction amount NL = 0. .2.

Rule: Confidence of Copy Number PS = 0.8 Confidence of Toner Density ZO = 0.8 Confidence of Humidity ZO = 0.8 Therefore, the assertion of the rule is the confidence of toner density correction amount NS = 0. Set to 8.

Rule: Confidence of Copy Number PS = 0.8 Confidence of Toner Density PL = 0.2 Confidence of Humidity NL = 0.2 Therefore, the assertion of the rule is the confidence of toner density correction amount NL = 0. .2.

Rule: Confidence of the number of copies PS = 0.8 Confidence of toner density PL = 0.2 Confidence of humidity ZO = 0.8 Therefore, the assertion of the rule is that the confidence of the toner density correction amount NS = 0. .2.

Next, each state quantity of the membership function of the toner density correction quantity is truncated according to the assertion result of the rules ~, and the overlapping portion is shown by the diagonal lines (see FIG. 12). The center of gravity of the shaded area is the control amount, and in this case, the toner density correction amount is -1.9%.

Although the control amount is calculated using the min-max centroid method as described above in the present embodiment, the consequent part of the inference rule is defined as a constant instead of a fuzzy set. , Different methods of inference procedures may be used, such as a simplified inference method of calculating a control amount by weighted average and a functional inference method of defining a consequent part as a function. Further, the shape of the membership function, the number and contents of inference rules can be changed according to experience and experimental results.

(Neural Network) On the other hand, as a method for obtaining the toner density correction amount, a neural network may be used in addition to the above fuzzy inference. FIG.
3 is a control block diagram of the neural network,
The toner density correction unit 80 (corresponding to the correction unit 70 shown in FIG. 2) includes a preprocessing unit 81 and a neural network 82.
And the post-processing unit 83 and the control unit 8 for controlling these
4, CPU85 (corresponding to the CPU60 shown in FIG. 2)
It is composed of The pre-processing unit 81 uses the durability counter 5
5 information, humidity information based on sensors 52 and 53, A
Toner density information from the TDC sensor 54 is input and converted into parameters that can be input to the neural network 82.

As shown in FIG. 14, the neural network 82 has a three-layer structure of an input layer, an intermediate layer, and an output layer, and each unit is a unit of an adjacent layer other than the layer to which it belongs. It is connected via a connection load. The signal travels in only one direction and enters the combined unit. This neural network 82
Under each condition of toner concentration, humidity, and durability, each information is input to the input layer, the true toner concentration is measured, and the optimum toner concentration correction amount is given to the output layer as a teacher value for learning. In the actual control, the optimum toner density correction amount is output from the output layer to the post-processing unit 83 from various information input to the input layer, the post-processing unit 83 calculates the true toner density, and outputs it to the CPU 85. .

As shown in FIG. 15, each unit of the neural network 82 uses a multi-input, one-output element. The input value x (i) is weighted by the connection weight w (i), and the weighted input value w (i) x (i)
Is the sum X, and after being transformed by the response function f,
Is output. The output value (unit value) y is

[0053]

[Equation 1]

Is represented by By the way, the outline of a learning model of a general neural network is performed by the procedure shown in FIG. First, the connection weight w of each connection is initialized with a random value (step S21). Next, an ideal output with respect to the input signal is externally applied as a teacher signal to this coupling weight, and evaluation is performed by referring to the evaluation standard (step S22). Next, the value of the coupling load is adjusted based on the evaluation result (step S24), and the evaluation is performed again.
By repeating such a process, the optimum value is gradually approached.

Learning of the neural network is performed by the three-layer backpropagation method. As shown in FIG. 17, the output I (i) of the i-th input layer unit becomes the input of the j-th intermediate layer unit due to the load sum at the coupling load W (ji), and the threshold θ (j) and the output are standardized. The output H (j) of the intermediate layer unit j is determined by the function f to be converted, and the output H (j) of the intermediate layer unit j is determined by the coupling load V (kj).
Becomes the input of the output layer unit k by the sum of weights at
The output O of the output layer unit k according to the threshold γ (k) and the function f
Consider a model in which (k) is determined.

The input data given to the input layer is propagated to the intermediate layer and the output layer while performing weighting calculation (calculation of sum of products of coupling weight and unit value). The unit value of the output layer is compared with the teacher data corresponding to the input data,
The output layer-intermediate layer and intermediate layer-input layer coupling strengths are modified so that the error therebetween is reduced. By repeating this modification, the value output from the output layer approaches the teacher data. By repeating this series of operations sufficiently for all input data, it is possible to output the values indicated by the teacher data for various input data.

The unit value of each layer is

[0058]

[Equation 2]

H (j): Value of j-th unit of intermediate layer O (k): Value of k-th unit of output layer I (i): Value of i-th unit of input layer θ (j): Threshold value of j-th unit of the hidden layer γ (k): Threshold value of k-th unit of the output layer W (ji): Load of j-th unit of the hidden layer on i-th unit of the input layer V (kj) : The weight function f (X) of the kth unit of the output layer with respect to the jth unit of the intermediate layer is a non-linear continuous function called a sigmoid function, which determines the input / output characteristics of the unit, and the output value is 0. It is limited to the range of ≦ H (j), O (j), I (j) ≦ 1.

F (X) = 1 / {1 + exp (−2x / uO)} = {1 + tanh (x / uO)} / 2 uO: Parameter for determining the slope of the sigmoid function Coupling weight W (ji), V (kj) And thresholds θ (j), γ
The initial value of (k) is set by a small random number value.

The square of the difference between the output and the teacher signal, that is, the function of the square error is called an error function. The error function E (p) for a certain pattern p and the error Et for all patterns are expressed by the following equation.

[0062]

(Equation 3)

The state in which the error Et is minimized is regarded as the optimum network, and the coupling weight and the threshold value are corrected so that Et is minimized. The correction is done as follows. The coupling weight V (kj) connected to the unit k in the output layer is calculated from the difference between the learning signal T (k) of the learning pattern and the output O (k) of the output layer.
And the error δ with respect to the threshold γ (k) of the unit k in the output layer
(K) is calculated by the following equation.

Δ (k) = 2 × {T (k) -O (k)} × O (k)
× {1-O (k)} / uO error δ (k) and coupling load W (j) from the intermediate layer to the output layer
From i) and the output H (j) of the intermediate layer, an error δ (j) with respect to the coupling load W (ji) connected to the intermediate layer unit j and the threshold θ (j) of the intermediate layer unit j is obtained by the following equation.

[0065]

(Equation 4)

By adding the product of the error δ (k) in the output layer unit k and the output H (j) of the intermediate layer unit j and the constant α, the coupling coefficient V from the intermediate layer unit j to the output unit k is added. Correct (kj). Also, the error δ (k)
And the constant β are added to modify the threshold γ (k) of the output layer unit k. V (kj) = V (kj) + α × δ (k) × H (j) γ (k) = γ (k) + β × δ (k) The error δ (j) in the intermediate layer unit j and the input layer unit i
By adding the product of the output I (i) and the constant α, the coupling coefficient W (ji) connected from the input layer unit i to the intermediate layer unit j is corrected. Further, by adding the product of the error δ (j) and the constant β, the threshold value θ of the intermediate layer unit j
Correct (j).

W (ji) = W (ji) + α × δ (j) × I (i) θ (j) = θ (j) + β × δ (j) This calculation formula is applied to one input / output pattern p. By executing the above, the error E (p) is minimized, and by performing learning on all input patterns, the error function Et as a whole is minimized.

In this embodiment, a hierarchical neural network having three input layers, five intermediate layers, and one output layer is used, and learning is performed by the back propagation method described above. When performing learning calculation and output calculation of the neural network, it is necessary to standardize input / output data between 0 and 1. Therefore, the obtained experimental data is normalized to a value of 0 to 1 within the range of the minimum value and the maximum value.

In the present embodiment, the teacher data obtained by the experiment was standardized within the range of values shown in Table 3 below, and set as the output value of the neural network.

[0070]

[Table 3]

For example, certain teacher data is standardized as shown in Table 4 below.

[0072]

[Table 4]

The teacher data thus obtained by the experiment is standardized, and learning is performed using the standardized teacher data.
In this embodiment, the total number of teacher data is 120.
The number of individual units is repeatedly learned by a computer until the squared error becomes sufficiently small, and the coupling weight between units and the unit threshold are determined. After the learning is completed, the processing program of the part for calculating the output of the neural network having the determined connection weight and threshold value is incorporated in the control program ROM as a toner concentration correction amount estimation routine.

At the time of copying, the data obtained by various sensor outputs are standardized so as to match the input / output type of the neural network, and then taken as input values of the processing program of the neural network to calculate output values. The toner density correction amount is obtained by changing the output value from the standardized value to the actual value.

Further, without incorporating the processing program of the neural network into the control ROM, the output for the combination of inputs is calculated by the previously learned neural network, the calculated values are built into the data ROM as a data table, and the input is made at the time of copying. For the data, output data may be selected from the data table to obtain the toner density correction amount.

In this embodiment, the learning is performed by the back propagation method using the three-layer neural network. However, in addition to this, the number of layers of the neural network, the number of units, the threshold value of the unit, the response function A method of constructing an optimum neural network by changing parameters such as is also conceivable.

[Brief description of drawings]

FIG. 1 is an internal configuration diagram showing a digital copying machine according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a control circuit of the copying machine.

FIG. 3 is a graph showing a relationship between a toner adhesion amount on a photoconductor and an output voltage of an AIDC sensor.

FIG. 4 is a graph showing a relationship between toner concentration according to humidity and output voltage of an ATDC sensor.

FIG. 5 is a graph showing the relationship between the number of copies and the output voltage of the ATDC sensor.

FIG. 6 is a flowchart showing a toner density control procedure.

FIG. 7 is a flowchart showing a procedure for detecting a corrected toner density.

FIG. 8 is a chart showing a membership function (number of copies) in fuzzy inference.

FIG. 9 is a chart showing a membership function (toner concentration) in fuzzy inference.

FIG. 10 is a chart showing a membership function (humidity) in fuzzy inference.

FIG. 11 is a chart showing a membership function (toner density correction amount) in fuzzy inference.

FIG. 12 is a chart showing calculation of a control amount in fuzzy inference.

FIG. 13 is a block diagram showing a control circuit of a neural network according to another embodiment.

FIG. 14 is an explanatory diagram of a neural network.

FIG. 15 is an explanatory diagram showing input / output of each unit of the neural network.

FIG. 16 is a flowchart showing a learning procedure of the neural network.

FIG. 17 is an explanatory diagram showing learning of a neural network.

[Explanation of symbols]

 20 ... Image forming unit 21 ... Photosensitive drum 23 ... Developing device 24 ... Toner hopper 51 ... AIDC sensor 52 ... Humidity sensor 53 ... Temperature sensor 54 ... ATDC sensor 55 ... Endurance counter 60, 85 ... CPU 70, 80 ... Toner density correction unit 71 ... Grid voltage transformer 72 ... Development bias transformer

Claims (4)

[Claims] 1. An image forming apparatus for transferring a toner image formed on an image carrier by an electrophotographic process onto a sheet, wherein toner is supplied to the electrostatic latent image formed on the image carrier to form a toner image. A developing unit that forms a toner image, a first sensor that detects a toner adhesion amount of a test toner image formed on the image carrier under constant image forming conditions, and a first sensor based on a detection value of the first sensor. Replenishing means for replenishing the developing device with toner, a second sensor for detecting a toner concentration in the developing device, a third sensor for detecting a variation factor of a detection value of the second sensor, The toner concentration correction amount based on the detection value of the second sensor, the detection value of the third sensor, and the detection values of the second and third sensors is described by a membership function, and the antecedent part of the control rule is the second. And the detection value of the third sensor, The consequent part is a fuzzy inference type rule described by the toner density correction amount,
Toner density correction means that applies the detected values of the second and third sensors as input values to the antecedent part, determines the fuzzy output value of the consequent part by fuzzy inference, and uses this output value as the toner density correction amount, An image forming apparatus comprising: a control unit that controls an electrophotographic process based on the toner concentration corrected by the toner concentration correction unit. 2. An image forming apparatus for transferring a toner image formed on an image carrier by an electrophotographic process onto a sheet, wherein toner is supplied to the electrostatic latent image formed on the image carrier to provide a toner image. A developing unit that forms a toner image, a first sensor that detects a toner adhesion amount of a test toner image formed on the image carrier under constant image forming conditions, and a first sensor based on a detection value of the first sensor. Replenishing means for replenishing the developing device with toner, a second sensor for detecting a toner concentration in the developing device, a third sensor for detecting a variation factor of a detection value of the second sensor, The detected value of the second sensor and the detected value of the third sensor are used as input values, and the toner density correction value preset based on the detected value of the third sensor is used as a teacher value. Learn the fluctuation characteristics of the detected value And a toner density correction unit that obtains a toner density correction amount by using the neural network, and a control unit that controls the electrophotographic process based on the toner density corrected by the toner density correction unit. Image forming apparatus. 3. The image forming apparatus according to claim 1, wherein the second sensor detects at least one of a usage history and a humidity of the developing device. 4. The control means is a charging potential of an image carrier,
The image forming apparatus according to claim 1 or 2, wherein at least one of a developing bias voltage and an image exposure amount is controlled.