JPH07281898A - Information processor and its method, and control system and its method - Google Patents

Abstract

PURPOSE:To perform a proper inference by using plural kinds of inference. CONSTITUTION:By a case base inference process 30, case base inference is performed and its evaluation is performed by a case data learning process 35. By a knowledge base inference process 40, knowledge base inference is performed and its evaluation is performed by a knowledge data learning process 45. By a model base inference process 50, model base inference is performed and its evaluation is performed by a model data learning process 55. By a merging weight calculating process 22, merging weight is calculated on the basis of evaluation values obtained by the respective learning processes. On the basis of inference results obtained by the inference of the respective inference processes and the merging weight calculated by the merging weight calculating process 22, an optimum inference result is calculated by a merging process 23.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to an information processing apparatus and method, and a control system and method.

[0002]

2. Description of the Related Art As an information processing device, there is an information processing device to which artificial intelligence such as fuzzy technology or artificial neural network is applied. Some of these information processing devices include those that perform information processing by performing inference such as case-based reasoning, knowledge-based reasoning, model-based reasoning, and the like.

The case-based inference is to infer by selecting a corresponding case from past case data. Case-based reasoning cannot handle case data that does not exist in the past, and is not practical until sufficient and consistent case data is collected. Therefore, in case-based reasoning, case data is generated by interpolating past case data, or pseudo case data is generated by applying noise.

In the knowledge (rule) based inference, a designer creates a knowledge data base (production rule) in advance, and the inference is performed using the knowledge data base. Generally, in knowledge-based reasoning, production rules are expressed using crisp numbers, and in fuzzy reasoning, production rules are expressed using fuzzy numbers. Hereinafter, the fuzzy reasoning is referred to as knowledge-based reasoning. In knowledge-based reasoning, knowledge must be consistent in order to obtain proper inference results, and it is difficult to acquire such consistent knowledge (knowledge of knowledge). Therefore, in knowledge-based reasoning, knowledge is modified by learning using cases (teaching information).

In model-based inference, a designer creates a physical model of a target system (controlled object) in advance, and infers using the physical model. In model-based reasoning, specialized knowledge is required to create a physical model of the target system. Also, it is difficult to create an accurate physical model. Therefore, in model-based inference, a physical model of the target system is created based on past case data.

However, in such a conventional information processing apparatus, since information processing is performed by the inference method designed in advance by the designer, an optimum inference result may not be obtained depending on the target system. In case-based reasoning, the amount of case data that is registered in advance becomes enormous, which requires high hardware costs.
It takes processing time. In knowledge-based reasoning and model-based reasoning, it takes time to reach practicable performance.

[0007]

DISCLOSURE OF THE INVENTION It is an object of the present invention to provide an information processing apparatus and method capable of calculating a plurality of types of inference for input information and calculating optimum output information.

The present invention also provides an application of such an information processing device. That is, it is an object of the present invention to provide a control system and method including a target amount generation device capable of setting an optimum target amount for a control target in a control device.

First, an information processing apparatus and method according to the present invention will be described.

The information processing apparatus according to the present invention comprises a plurality of types of inference means for inferring input information according to separate data bases for controlling a controlled object or processing information, and the plurality of types of inference means. The output information is output based on the evaluation means for evaluating each of the obtained inference results, the inference results obtained by the inference means of the plurality of types, and the evaluation results obtained by the evaluation means. The output information calculating means for calculating is provided.

In the information processing method according to the present invention, a plurality of types of inferences are performed on input information according to separate data bases for controlling a control target or processing information, and a plurality of types of inferences obtained by the plurality of types of inferences are performed. Each of the inference results is evaluated, and output information is calculated based on the multiple inference results obtained by the multiple types of inference and the multiple evaluation results obtained by the evaluation.

In one embodiment of the present invention, the plurality of types of inferences are based on a knowledge data base, a knowledge base inference is performed, a case base inference is performed based on a case data base, and a model data base is inferred.
At least two of the model-based inferences that infer according to the base.

According to the present invention, a plurality of types of inference are performed on input information according to different databases. An evaluation is performed on each of a plurality of inference results obtained by a plurality of types of inference. Output information is calculated based on a plurality of inference results and a plurality of evaluation results.

Therefore, since a plurality of types of inference are performed on the input information at the same time and the evaluation is performed,
Even if one inference cannot obtain a proper inference result, other inference can obtain an appropriate inference result. As a result, the optimum output information for the input information can be calculated based on the proper inference result obtained by any inference.

In another embodiment of the present invention, the output information calculation unit fuses the plurality of evaluation results obtained by the evaluation unit by normalizing each evaluation result with the maximum value of the evaluation results. Inference for calculating output information based on fusion weight calculation means for calculating respective weights, a plurality of inference results obtained by the plurality of types of inference means, and a plurality of fusion weights calculated by the fusion weight calculation means It is realized by means of result fusion.

According to this embodiment, with respect to a plurality of evaluation results obtained by the evaluation, the fusion weights are calculated by normalizing the respective evaluation results by the maximum value of the evaluation results. Output information is calculated based on a plurality of inference results and a plurality of fusion weights.

Therefore, the output information can be calculated based on the fusion weight based on the evaluation result and the inference result.

In this embodiment, the fusion weight calculating means sets the fusion weight to 0 when the calculated fusion weight is smaller than a predetermined rejection threshold.

Therefore, since the fusion weight is smaller than the rejection threshold, that is, the inference result having a small evaluation result before being normalized is not considered in the calculation of the output information,
The output information can be calculated using other appropriate inference results.

In this embodiment, the fusion weight calculating means sets the fusion weight other than the fusion weight to 0 when there is a fusion weight equal to or higher than a predetermined priority adoption threshold among the calculated fusion weights. Is.

Therefore, since the output information is calculated using the inference result in which the fusion weight is equal to or higher than the priority threshold, that is, the evaluation result before being normalized is large, other improper inference results should be considered. Instead, the optimum output information can be obtained.

In this embodiment, preferably, the inference result fusion means calculates a distance between one inference result and another inference result with respect to a plurality of inference results obtained by the plurality of types of inference means, respectively. A modified inference result obtained by modifying the one inference result is calculated based on the other inference result and the distance thereof, and the modified inference result is calculated based on the plurality of fusion weights calculated by the fusion weight means and the plurality of distances. The output information is calculated by calculating the reliability of each of the plurality of modified inference results and selecting the modified inference result that maximizes the reliability.

According to this embodiment, the distance between one inference result and another inference result is calculated for each of a plurality of inference results obtained by a plurality of types of inference. A modified inference result in which one inference result is modified is calculated based on another inference result and its distance. The reliability of the modified inference result is calculated based on the plurality of fusion weights and the plurality of distances. The output information is calculated by selecting the modified inference result that maximizes the reliability.

Therefore, the output information is calculated by selecting one of the modified inference results which has been modified in consideration of the other inference results with respect to one inference output result and which has the maximum reliability. Therefore, highly reliable output information can be obtained.

In another embodiment of the present invention, there is further provided a learning means for correcting each of the separate data bases by learning based on the inputted teacher information.

Therefore, since a separate database used for each inference is generated by learning, each inference can make an inference using an optimal data base. This makes it possible to obtain accurate inference results.

Next, a control system and method according to the present invention will be described.

The control system according to the present invention comprises a first sensor for measuring the controlled variable of the controlled object, an actuator for operating the controlled object based on a given manipulated variable, and a controlled variable measured by the first sensor. In a control system including a control device that determines an operation amount such that the target amount approaches a given target amount, the target amount generation device that sets the target amount in the control device is controlled by the target information regarding the control target. Second sensor for measuring
Target information measured by the sensor and input information to be input are inferred according to separate data bases for generating target amounts, and a plurality of types of inference means for calculating target amounts, respectively, the plurality of types described above. Evaluating means for evaluating each of a plurality of target quantities obtained by the inference means theory and calculating an evaluation value,
And target quantity fusion means for calculating an optimum target quantity for the controlled object based on a plurality of target quantities obtained by the plurality of types of inference means and a plurality of evaluation values obtained by the evaluation means, The optimum target amount calculated by the target amount fusion means is given to the control device.

The control method according to the present invention includes a first sensor for measuring the controlled variable of the controlled object, an actuator for operating the controlled object based on a given manipulated variable, and a controlled variable measured by the first sensor. Determines the manipulated variable so that it approaches a given target amount, and in a control system equipped with a control device for giving to the actuator, measures the target information relating to the control target, and measures the measured target information and the input information to be input. For each of the above, multiple types of inference are performed according to separate data bases for generating target amounts, target amounts are calculated, and multiple target amounts obtained by the above multiple types of inference are evaluated and evaluated. The values are calculated respectively, and the target values obtained by the above-mentioned plurality of types of inference and the plurality of evaluation values obtained by the above evaluation are obtained. Zui calculates an optimum target amount for the controlled system, the optimum target amount calculated is intended to provide to the control device.

In one embodiment of the present invention, the plurality of types of inferences are based on a knowledge data base, a knowledge base inference is performed, a case data inference is performed based on a case data base, and a model data inference is performed.
At least two of the model-based inferences that infer according to the base.

According to the present invention, the object information regarding the controlled object is measured. Multiple types of inference are performed on the measured target information and the input information to be input according to separate databases for generating the target quantity.
An evaluation is carried out for each of a plurality of target quantities obtained by a plurality of types of inference. An optimum target amount for the control target is calculated based on a plurality of target amounts obtained by a plurality of types of inference and a plurality of evaluation values obtained by the evaluation. The calculated optimum target amount is given to the control device.

Therefore, since a plurality of types of inferences are simultaneously performed and evaluated with respect to the target information and the input information regarding the controlled object, even if one inference may not be able to obtain an appropriate target amount, A proper target amount can be obtained by reasoning. This makes it possible to calculate the optimum target amount for the controlled object based on the appropriate target amount obtained by any inference. Therefore, the control device can stably control the controlled object by the optimum target amount.

In a preferred embodiment of this invention,
The target amount calculation means calculates fusion weights by normalizing each evaluation value of the plurality of evaluation values obtained by the evaluation means by the maximum evaluation value, and the plurality of fusion weight calculation means. It is realized by the target amount fusion means for calculating the optimum target amount based on the plurality of target amounts obtained by the inference means of the types and the plurality of fusion weights calculated by the fusion weight calculation means.

According to this embodiment, with respect to the plurality of evaluation values obtained by the evaluation, the fusion weights are calculated by normalizing the respective evaluation values by the maximum value of the evaluation values. The optimum target amount is calculated based on the plurality of target amounts and the plurality of fusion weights.

Therefore, it is possible to calculate the optimum target amount for the above-mentioned controlled object based on the fusion weight and the target amount based on the evaluation value.

In this embodiment, the fusion weight calculation means sets the fusion weight to 0 when the calculated fusion weight is smaller than a predetermined rejection threshold.

Therefore, a target amount whose fusion weight is smaller than the rejection threshold, that is, a target amount whose evaluation value before being normalized is small, is not considered in the calculation of the optimum target amount, and therefore another appropriate target amount is used. Therefore, the optimum target amount can be calculated.

In this embodiment, the fusion weight calculation means sets fusion weights other than the fusion weight to 0 when there is a fusion weight equal to or higher than a predetermined priority adoption threshold among the calculated fusion weights. Is.

Therefore, since the optimal target amount is calculated using the target amount whose fusion weight is equal to or higher than the priority threshold, that is, the target amount having a large evaluation value before being normalized, other inappropriate target amounts are calculated. The optimum target amount can be obtained without considering

In this embodiment, preferably, the target amount fusion means calculates a distance between one target amount and another target amount for each of the plurality of target amounts obtained by the plurality of types of inference means, Based on the other target amounts and the distances thereof, modified target amounts obtained by modifying the one target amount are respectively calculated, and based on the plurality of fusion weights calculated by the fusion weight means and the plurality of distances, The optimum target amount is calculated by calculating the reliability of each of the plurality of correction target amounts and selecting the correction target amount that maximizes the reliability.

According to this embodiment, the distance between one target amount and another target amount is calculated for each of a plurality of target amounts obtained by a plurality of types of inference. Based on the other targets and the distances thereof, the modified target amounts obtained by modifying the one target amount are calculated. The reliability of the correction target amount is calculated based on the plurality of fusion weights and the plurality of distances. The optimum target amount is calculated by selecting the correction target amount that maximizes reliability.

Therefore, the optimum target amount is calculated by selecting the one that has the maximum reliability from the corrected target amounts that have been corrected in consideration of the other target amounts for one target amount. Therefore, it is possible to obtain an optimal target amount with high reliability.

In another embodiment of the present invention, the target amount generation device further comprises learning means for modifying each of the separate data bases by learning based on the inputted teacher information.

Therefore, since a separate database used for each inference is generated by learning, each inference can make an inference using an optimum data base. This makes it possible to obtain an accurate target amount.

In another embodiment of the present invention, the target amount generating device sets a preset basic target amount pattern for the control amount to quickly settle to the target amount and to prevent an overshoot amount from occurring. The control device further includes a target amount pattern creating unit for creating a target amount pattern by normalizing using the optimum target amount calculated by the calculating unit, and the target amount pattern created by the target amount pattern creating unit. To give to.

According to this embodiment, the preset basic target amount pattern for the control amount to settle to the target amount quickly and the overshoot amount does not occur is standardized by using the optimum target amount. A target quantity pattern is created. The created target amount pattern is given to the control device.

Therefore, the control device can set the control amount to the target amount quickly and prevent the overshoot amount from being generated by giving the created target amount pattern.
The controlled object can be controlled stably.

The control device according to the present invention controls the extrusion speed in the extruder.

Stable extrusion molding can be performed by the calculated optimum target amount (preferably the target amount pattern), that is, the optimum target extrusion speed.

[0050]

【Example】

 Table of contents 1 System configuration 2 Detailed configuration of target value generator 2.1 Outline of target extrusion velocity pattern generation 2.2 Case data 2.3 Case-based reasoning 2.3.1 Case-based reasoning processing 2.3.2 Case-data learning processing 2.4 Knowledge-based reasoning 2.4.1 Knowledge Base inference processing 2.4.2 Knowledge data learning processing 2.5 Knowledge base inference processing 2.5.1 Model-based inference processing 2.5.2 Model data learning processing 2.6 Fusion weight calculation 2.7 Fusion of inference results 2.8 Creation of target extrusion speed pattern 2.9 Target extrusion Speed pattern setting

1 System configuration

FIG. 1 is a diagram showing the overall configuration of a control system for an extruder.

A buret is set on the buret setting machine 2. The molten aluminum is contained in the buret. A die 3 which is a mold of an aluminum sash is attached to one end of the buret. The dice 3 can be exchanged.

FIG. 2 shows an example of the die 3. In the cross section of this die (shown by hatching), the shorter length is called the minor diameter, and the longer length is called the major diameter.

The inside of the buret is pressurized by the extrusion pump 1. When pressurized by the extrusion pump 1, the molten aluminum in the buret is extruded from the die 3. Aluminum sash 4 extruded from die 3
Is cooled while being pulled by the tensioner 6. In this way, the aluminum sash is formed.

The shape of the aluminum sash 4 formed is determined by the shape of the die 3 which is a mold. By exchanging the die 3, an aluminum sash having a desired shape can be formed. There are thousands of types of aluminum sashes to be formed, and there is a die for each aluminum sash.

The extrusion speed of aluminum extruded from the die 3 is controlled as follows.

The extrusion speed of aluminum extruded from the die 3 is measured by a speed sensor (not shown).
The measured extrusion speed measured by the speed sensor is given to the control device 11.

The control device 11 determines the extrusion speed command value so that the measured extrusion speed measured by the speed sensor approaches the target extrusion speed given from the target value generation device 10 described later. The extrusion speed command value determined by the control device 11 is given to the extrusion pump 1.

The extrusion pump 1 pressurizes the inside of the buret with a pressure based on the extrusion speed command value given from the control device 11. Aluminum will be extruded at a rate according to the pressure applied in the buret.

In this way, the extrusion speed is controlled.

The quality of the formed aluminum sash is determined by the extrusion speed of the aluminum extruded from the die 3. Each die has its own unique extrusion speed (hereinafter referred to as "ideal extrusion speed") depending on its shape. In order to form a high quality aluminum sash, the extrusion rate of aluminum extruded from the die 3 (measured extrusion rate) must be kept constant at the ideal extrusion rate (target extrusion rate).

When the die 3 is replaced, the ideal extrusion speed of the die is set in the controller 11 as the target extrusion speed. As described above, since there are thousands of types of dies and each die has an ideal extrusion speed, it is not possible to register all the ideal extrusion speeds in the data base in advance. Therefore, when the operator replaces the die 3, the operator must set the ideal extrusion speed corresponding to the shape of the die 3 as the target extrusion speed in the controller 14 (ideal extrusion speed setting operation).

Further, at the start of extrusion molding, it may take time until the measured extrusion speed reaches the target extrusion speed (ideal extrusion speed). In addition, the measured extrusion speed may have a large overshoot. Therefore, the operator can set the measured extrusion speed to the ideal extrusion speed quickly, and
A target extrusion speed pattern that does not cause an overshoot amount must be set in the control device 11 (target extrusion speed pattern setting operation).

As described above, the operator must perform two operations such as an ideal extrusion speed setting operation and a target extrusion speed pattern setting operation every time the die is replaced. In order to reduce such operator's operation and to form a higher quality aluminum sash, the target value generation device 10 generates an optimum target extrusion speed pattern and sets it in the control device 11.

As will be described in detail later, the target value generator 10 obtains by performing image processing on the die number entered by the operator and the image information obtained by photographing the die 3 attached to the buret. The target extrusion speed pattern is generated based on the die shape information.

The camera 12 photographs the dice 3 attached to the buret. Image information obtained by photographing the dice 3 is given from the camera 12 to the image processing device 13.

The image processing device 13 creates the die shape information regarding the dice based on the image information given from the camera 12.

The image processing device 13 performs image processing on the image information given from the camera 12 to extract the minor axis and major axis of the die cross section and the cross sectional area.

The minor axis and major axis of the die section are the lengths shown in FIG.

The cross-sectional area of the die is the area of the hatched portion in the die shown in FIG.

The image processing device 13 creates die shape information based on the minor and major diameters of the die cross section and the cross sectional area.

The die shape information is created when the die is exchanged and when the target extrusion speed pattern generated by the other target value generation device 10 is generated. The die shape information may be created for each extrusion molding.

The die shape information S includes roundness S1 and thickness S
2 and complexity S3.

The roundness S1 is represented by the ratio of the short diameter to the long diameter of the die section. That is, the roundness S1 is
It is represented by "major axis / minor axis".

The thickness S2 is represented by the cross-sectional area of the die.

The complexity S3 is represented by the ratio of the area of the circumscribed rectangle of the die cross section, that is, "major axis × minor axis" to the sectional area. That is, the complexity S3 is represented by "cross-sectional area / (major axis x minor axis)".

Hereinafter, the die shape information is represented by S (S1, S2,
It is represented by S3).

Die shape information S (S1, S2, S3)
Is given from the image processing device 13 to the target value generation device 10.

The input device 14 is used by the operator to input teacher information.

The teacher information includes a die number Dn, a manual target extrusion speed pattern VP, a teacher ideal extrusion speed TVn and quality evaluation information C.

The die number Dn represents the number of the die attached to the buret. Dice number D
When the die attached to the buret is replaced, n is the number of the die entered by the operator.

The manual target extrusion speed pattern VP is input when the operator determines that the target extrusion speed pattern generated by the target value generator 10 is incorrect.

As shown in FIG. 3, the manual target extrusion speed pattern VP includes the number of speed changes N, target extrusion speeds V0 and V1.
And V2, and speed change times T0, T2 and T
It is defined by 3.

The speed change number N represents the number of times the target extrusion speed changes. In the target extrusion speed pattern shown in FIG. 3, the target extrusion speed changes twice from the target extrusion speed V0 to V1 and from V1 to V2, so the number of speed changes N is "2".

The target extrusion speed V0 is set higher than the ideal extrusion speed so that the measured extrusion speed can settle to the ideal extrusion speed quickly.

The speed change time T0 represents the time for switching the target extrusion speed V0 to V1.

The target extrusion speed V1 is set to be lower than the ideal extrusion speed so that the measured extrusion speed does not overshoot.

The speed change time T1 represents the time for switching the target extrusion speed V1 to V2.

The target extrusion speed V2 is the ideal extrusion speed of the die described above. This target extrusion speed V2 is equal to a teacher ideal extrusion speed TVn described later.

The speed change time T2 represents the end time of one extrusion molding. This speed change time T2 is 1
It is equal to the time required to perform one extrusion, that is, the extrusion time.

When inputting the manual target extrusion speed pattern VP, the operator inputs these parameters (speed change number N, target extrusion speeds V0, V1 and V3, and speed change times T0, T1 and T2).

Below, the manual target extrusion speed pattern is set to VP.
(N, V0, V1, V2, T0, T1, T2).

The teacher's ideal extrusion rate TVn is the ideal extrusion rate for the die attached to the buret. The teacher's ideal extrusion speed TVn is input when using a die that has never been used in the past. Further, since the target extrusion speed V2 of the above-mentioned manual target extrusion speed VP is equal to the teacher ideal extrusion speed TVn, it is not necessary to input the teacher ideal extrusion speed TVn when the manual target extrusion speed is input.

The quality evaluation information C is a quality evaluation value when the operator evaluates the formed aluminum sash.

The quality evaluation information C includes the thickness unevenness C1 and the sash deformation C2. The wall thickness unevenness C1 and the sash deformation C2 are represented by seven quality evaluation values, as shown in FIGS. 4 (A) and 4 (B), respectively.

As shown in FIG. 4 (A), the wall thickness unevenness C1 is "no wall thickness unevenness", "slight wall thickness unevenness", "slight wall thickness unevenness", "wall thickness unevenness". There is uneven thickness ”,
There are 7 levels of "Slightly uneven wall thickness", "Large uneven wall thickness" and "Very large uneven wall thickness", respectively "0", "1", "2", "3", " It is represented by quality evaluation values of "4", "5" and "6".

As shown in FIG. 4 (B), the sash deformation C2 has "no sash deformation", "slight sash deformation", "some sash deformation", and "sash deformation". , "Sash deformation is slightly large", "Sash deformation is large" and "Sash deformation is very large", and there are 7 stages, "0", "1", "2", "3", respectively.
It is represented by quality evaluation values of "4", "5" and "6".

The operator inputs, as the quality evaluation information C, any one of the quality evaluation values from "0" to "6" for the wall thickness unevenness C1 and the sash deformation C2.

Hereinafter, the quality evaluation information is represented by C (C1, C2).

2 Detailed Configuration of Target Value Generating Device

The target value generation device 10 calculates the ideal extrusion speed by making a plurality of types of inference with respect to the die number input from the input device 14 and the die shape information given from the image processing device 13, and calculates the ideal extrusion speed. To produce the target extrusion rate pattern.

In this embodiment, the plurality of types of inference are knowledge-based inference, case-based inference and model-based inference.

The knowledge base inference is fuzzy inference in this embodiment. In general, knowledge-based reasoning is one in which the antecedent and consequent variables of a production rule of knowledge data are expressed using crisp numbers, and in fuzzy reasoning the antecedent and consequent variables are fuzzy numbers. Is expressed using. Knowledge-based reasoning and fuzzy reasoning may be performed separately. Other inferences may be used in addition to the above three inferences.

FIG. 5 is a functional block diagram showing the detailed structure of the target value generator 10.

The target value generator 10 can be realized by a programmed computer system. Further, the target value generating apparatus 10 can be realized entirely by a hardware architecture, a part of it by hardware, and a part of it by software.

The target value generation device 10 uses the input / output history memory 2
1, fusion weight calculation processing 22, fusion processing 23, target extrusion speed pattern creation processing 24, target extrusion speed pattern memory 25,
Target extrusion speed pattern setting process 26, case-based reasoning process
30, case data learning processing 35, knowledge base inference processing 40, knowledge data learning processing 45, model base inference processing 50 and model data learning processing 55. Each of these processes is a program routine,
It is realized by the system and the software running on it. The input / output history memory 21 and the target extrusion speed pattern memory 25 are both realized by a disk memory or the like.

2.1 Outline of Target Extrusion Rate Pattern Generation

Details of each process constituting the target value generation device 10 will be described later, but the target extrusion speed pattern generation in the target value generation device 10 will be briefly described below.

FIGS. 6 and 7 are flow charts showing the process of generating the target extrusion speed pattern in the target value generation device 10.

When the die attached to the buret is exchanged by the operator, the die number Dn of the exchanged die is input from the input device 14 to the target value generator 10.

Also, the dice exchanged by the operator is photographed by the camera 12. The image information obtained by the camera 12 is subjected to image processing by the image processing device 13 so that the die shape information S (S
1, S2, S3) are created. The die shape information S (S1, S2, S3) created by the image processing device 13 is given to the target value generating device 10.

The target value generator 10 takes in the die number Dn input from the input device 14 (FIG. 6; step 101).
).

The target value generator 10 takes in the die shape information given from the image processor 13 (FIG. 6; step 10).
2).

With respect to the taken die number Dn, case-based reasoning is performed by the case-based reasoning process 30 to obtain the ideal extrusion speed IV1 (FIG. 6; step 103).

The ideal extrusion speed IV1 obtained by the case-based reasoning in the case-based reasoning process 30 is evaluated by the case data learning process 35, and the evaluation value ES is obtained.
1 is obtained (FIG. 6; step 104).

Die shape information S (S1, S2 taken in)
, S3), knowledge base inference is performed by the knowledge base inference processing 40 to obtain the ideal extrusion rate IV2 (FIG. 6; step 105).

The ideal extrusion rate IV2 obtained by the knowledge base inference in the knowledge base inference processing 40 is evaluated by the knowledge data learning processing 45, and the evaluation value ES
2 is obtained (FIG. 6; step 106).

Die shape information S (S1, S2 taken in)
, S3), model-based inference processing 50 performs model-based inference to obtain an ideal extrusion rate IV3 (FIG. 6; step 107).

The ideal extrusion speed IV3 obtained by the model-based inference processing 50 is evaluated by the model-data learning processing 55 to obtain an evaluation value ES3 (FIG. 6; step 108).

The ideal extrusion rate IV1 obtained by inference in each inference process is based on the evaluation values ES1, ES2 and ES3 obtained in the evaluation in each learning process.
, IV2 and IV3 respectively, the fusion weights WT1, WT2 and WT3 are fused weight calculation processing.
22 respectively (FIG. 6; step 109)
).

The ideal extrusion speeds IV1, IV2 and IV3 obtained by inference in each inference process, and the fusion weight WT1 calculated by the fusion weight calculation process 22,
Based on WT2 and WT3, the optimum ideal extrusion rate IV for the exchanged die is calculated by the fusion process 23 (FIG. 6; step 110).

By using the ideal extrusion speed IV calculated by the fusion process 23, the measured extrusion speed is settled to the ideal extrusion speed quickly, and the basic target extrusion speed pattern that does not cause an overshoot is standardized. The target extrusion speed pattern is created by the target extrusion speed pattern creation processing 24 (FIG. 6; step 111).

The target extrusion speed pattern created by the target extrusion speed pattern creation processing 24 is displayed on the display device (not shown) (FIG. 7; step 112).

The operator determines whether or not the displayed target extrusion speed pattern is appropriate. If the operator determines that the target extrusion speed pattern is appropriate, the operator selects "OK".

When "OK" is selected (in step 113,
YES), the target extrusion speed pattern created by the target extrusion speed pattern creation processing 24 is output from the target extrusion speed pattern setting processing 26 to the control device 11 (FIG. 7; step 119). The controller 11 will perform extrusion molding using this target extrusion speed pattern.

When the operator judges that the operator is not appropriate (NO in step 113), "Please input the manual target extrusion speed pattern" is displayed on the display device (not shown) (FIG. 7; step 114). .

From the input device 14, the operator operates the manual target extrusion speed pattern VP (N, V0, V1, V2, T0, T
Enter 1, T2).

Manual target extrusion speed pattern VP (N, V0, V1, V2, T0, T1,
T2) is taken into the target extrusion speed pattern setting process 26 (FIG. 7; step 115), and the taken-in manual target extrusion speed pattern is outputted from the target extrusion speed pattern setting process 26 to the control device 11 (FIG. 7). ; Step 116). Controller 11
Will perform extrusion using this manual target extrusion rate pattern.

The operator judges the quality of the aluminum sash molded by using the inputted manual target extrusion speed pattern. When the operator determines that the input manual extrusion speed pattern is appropriate, the operator causes the target value generation device 10 to register the target manual extrusion speed pattern.

When "Register" is selected (Y in step 117)
ES), manual target extrusion speed pattern is case data inference processing
It is registered in 30 (FIG. 7; step 118).

When the operator determines that the input manual extrusion speed pattern is incorrect (NO in step 117),
Returning to step 115, the new manual target extrusion speed pattern input by the operator will be captured.

In this way, when the die attached to the buret is replaced, the target extrusion speed pattern is generated by the target value generation device 10 for the replaced die. This target extrusion speed pattern is set in the control device 11.

2.2 Case data

The information (data) input (given) to the target value generation device 10, the information (data) generated in the target value generation device 10, etc. are treated as case data every time the target extrusion speed pattern is generated. It is stored in the input / output history memory 21. The information (data) stored in the input / output history memory 21 is collectively referred to as “history information”.

FIG. 8 shows an example of the format of the case data stored in the input / output history memory 21.

The input / output history memory 21 is used by the target value generator 10
The data on the target extrusion speed pattern generation performed at the k-th time in the above is stored as the case number k.

In the example shown in FIG. 8, the case number is k = 822.

The information (data) input to the target value generator 10 stored in the input / output history memory 21 is as follows.

When the die number Dn is input from the input device 14, the die number is stored as Dn, k.

From the image processing device 13, the die shape information S (S
1, S2, S3), the dice information is S
It is stored as k (S1, k, S2, k, S3, k).

When the manual target extrusion speed pattern VP (N, V0, V1, V2, T0, T1, T2) is input from the input device 14, the manual target extrusion speed pattern is V
Pk (Nk, V0, k, V1, k, V2, k, T0, k, T1, k
, T2, k) in the input / output history memory 21.
At this time, the target extrusion speed V2 is the teacher's ideal extrusion speed TVn,
It is stored in the input / output history memory 21 as k, and the speed change time T2 is stored as the extrusion time TK.

When the teacher ideal extrusion speed TVn is input from the input device 14, it is stored in the input / output history memory 21 as the teacher ideal extrusion speed TVn, k.

From the input device 14, quality evaluation information C (C1, C
2) is entered, the quality evaluation information is Ck.
It is stored as (C1, k, C2, k).

The information (data) generated in the target value generation device 10 and stored in the input / output history memory 21 are as follows.

The inference result obtained by the inference in each inference process, that is, the ideal extrusion speed IV1, k obtained by the case-based inference in the case-based inference process 30,
The ideal extrusion rate IV2, k obtained by the knowledge-based reasoning in the knowledge-based reasoning processing 40 and the ideal extrusion rate IV3, k obtained by the model-based reasoning in the model-based reasoning processing 50 are stored.

With respect to the inference results obtained by each inference, the fusion weights WT1, k, WT2, k and WT3, k calculated by the fusion weight calculation process 22 on the basis of the evaluation results obtained by the evaluation in each learning process. Is memorized.

The ideal extrusion speed IVk calculated by the fusion calculation processing 23 is stored.

2.3 Case-Based Reasoning

The case-based reasoning process 30 is a process for carrying out case-based reasoning for the die number Dn, k input from the input device 14 to determine the ideal extrusion speed IV1, k.

The case data learning processing 35 evaluates the ideal extrusion speed IV1, k obtained by the case-based reasoning in the case-based reasoning processing 30 to calculate the evaluation value ES1, k, and the case-based reasoning processing 30 The case data to be used is learned.

FIG. 9 is a functional block diagram showing the detailed configurations of the case-based reasoning process 30 and the case data learning process 35. The case-based reasoning process 30 is a case data base 31.
And case-based reasoning operation processing 32. The case data learning process 35 includes a case base inference output history memory 36, a case base inference evaluation process 37, and a case data correction process 38. Each of these processes is a program routine.

2.3.1 Case-based reasoning processing

The case data base 31 stores, for each die, a die number Dn and a teacher ideal extrusion speed TVn for the die as case data.

In the input / output history memory 21, the die number Dn,
Since k and the teacher ideal extrusion speed TVn, k are stored, the input / output history memory 21 may be used instead of the case data base 31.

The case-based inference operation processing 32 is performed by the input device 14
For the die number Dn, k input from, the case data base 31 is searched, and the teacher ideal extrusion speed TVn, k corresponding to the die number Dn, k is read as the ideal extrusion speed IV1, k. It is reasoning.

When a plurality of teacher ideal extrusion speeds are stored for the same die number, the case-based inference calculation processing 32 reads the latest teacher ideal extrusion speed as the ideal extrusion speed. In the case-based reasoning calculation process 32, the ideal extrusion speed corresponding to the die number Dn, k is the case data base 31.
When it is not, the ideal extrusion speed IV1, k is not output.

The ideal extrusion speed IV1, k obtained by the case-based reasoning in the case-based reasoning operation processing 32 is given to the fusion processing 22, the case-based reasoning evaluation processing 37, and the case data correction processing 38, and also the input / output history memory. 21 and the case-based reasoning output history memory 36.

2.3.2 Case data learning processing

The case-based inference output history memory 36 stores the ideal extrusion speed V1, given by the case-based inference operation process 32.
k is stored for each case-based reasoning (case number) in the case-based reasoning operation processing 32.

Since the ideal extrusion speed IV1, k obtained by the case-based reasoning is stored in the input / output history memory 21, the case-based reasoning output history memory 36 may be omitted.

The case-based reasoning evaluation processing 37 evaluates the ideal extrusion speed IV1, k obtained by the case-based reasoning to calculate the evaluation value ES1, k.

The evaluation items evaluated by the case-based evaluation processing 37 are consistency of past cases. The contradiction of the past cases becomes the evaluation value ES1, k.

In the past case consistency ES1, k, the difference (distance) between the two die shape information is small at any two points in the past, and is 2 obtained by the case-based reasoning.
The two ideal extrusion rates do not differ greatly. That is, the dies have similar shapes and the ideal extrusion speeds do not differ greatly. This takes advantage of the fact that similar shaped dies have similar ideal extrusion rates.

The consistency ES1, k of the past case is obtained by calculating the past time points n and m (n ≠ n stored in the input / output history memory 21).
m) (n, m represents a past time (case number)) die shape information Sn (roundness S1, n, thickness S2, n, complexity S3, n) and Sm (roundness S1, m) , Thickness S2, m, complexity S3, m), and case-based reasoning output history memory
It is calculated by the equation (1) based on the ideal extrusion speeds IV1, n and IV1, m stored in 36 (or the input / output history memory 21).

[0166]

[Equation 1]

Here, n and m are the input / output history memory 21.
And all case numbers stored in the case-based reasoning output history memory 36. M is a zero-crossing prevention margin for preventing the denominator of the equation (1) from becoming 0.

The evaluation value ES1, k evaluated by the case-based inference evaluation process 37 (calculated by the equation (1)) is
This is given to the fusion weight calculation processing 22.

The case data correction processing 38 is for learning the case data stored in the case data base 31.

When the manual target extrusion speed pattern is input from the input device 14, the target extrusion speed V1, k of the manual extrusion speed pattern becomes the teacher ideal extrusion speed TVn, k. k (target extrusion speed V2,
Case data is learned using k).

In the case data correction processing 38, the die number Dn, k input from the input device 14 and the target ideal extrusion speed V are input.
Register 2, k as case data in the case database 31. This is the learning of case data in the case data correction processing 38.

Further, when a die which has never been used in the past is attached to the buret, the ideal extrusion speed of the die is not registered in the case data base 31. At this time, the operator may input the teacher's ideal extrusion speed TVn of the die from the input device 14. The case data correction processing 38 learns the case data using the input teacher ideal extrusion speed TVn.

As described above, the case-based inference is performed on the input die number, the inference result by the case-based inference is evaluated, and the case data is learned.

2.4 Knowledge-based reasoning

The knowledge base inference processing 40 is performed by the image processing device 13
With respect to the die shape information Sk given from, the knowledge base inference is performed using the knowledge data to obtain the ideal extrusion speed IV2,
It determines k.

The knowledge data learning process 45 evaluates the ideal extrusion speed IV2, k obtained by the knowledge base inference in the knowledge base inference process 40 to calculate the evaluation value ES2, k, and inputs from the input device 14. The learning data is used to modify the knowledge data used for the knowledge base inference using the teaching information.

As described above, in this embodiment, the knowledge base inference is fuzzy inference.

FIG. 10 is a functional block diagram showing the detailed configurations of the knowledge base inference processing 40 and the knowledge data learning processing 45. The knowledge base inference processing 40 is based on the knowledge database 41
And a knowledge base inference operation processing 42. The knowledge data learning process 45 includes a knowledge base inference output history memory 46, a knowledge base inference evaluation process 47, and a knowledge data correction process 48. Each of these processes is a program routine.

2.4.1 Knowledge-based inference processing

The knowledge data base 41 stores the knowledge data used by the knowledge base inference operation processing 42 for the knowledge base inference.

An example of the production rules (fuzzy rules) for the knowledge data stored in the knowledge data base 41 is shown below.

R1: If the die shape is round (because it is difficult to cool), slow down the extrusion rate R2: If the die shape is flat (because it is easy to cool), increase the extrusion rate R3: If the die shape is If it is thin (because it is easily deformed), the extrusion speed is slowed down a little. R4: If the die shape is thick (because it is difficult to deform), the extrusion rate is made a little faster. ) Decrease the extrusion speed a little R6: If the die shape is complicated (because it is difficult to deform), increase the extrusion speed a little .....

FIG. 11 shows an example of the membership function of the antecedent part variable and the consequent part variable of the knowledge data.

FIG. 11A shows an example of two membership functions representing the language information "round" and "flat" regarding the antecedent variable "roundness".

FIG. 11B is an example of two membership functions representing the language information “thin” and “thick” with respect to the antecedent variable “thickness”.

FIG. 11C is an example of two membership functions representing the language information “complex” and “simple” with respect to the antecedent variable “complexity”.

FIG. 11 (D) shows that the consequent part variable “extrusion speed” is 1 for language information “slow” for each rule.
Two, "Slightly slow", "Slightly fast"
It is an example of a singleton for the two, and one for the "speed up".

The knowledge base inference operation processing 42 uses the die shape information Sk (roundness S1, k,
For the thickness S2, k and the complexity S3, k, the knowledge data stored in the knowledge data base 41 is used to perform knowledge base inference (fuzzy inference) to calculate the extrusion speed.

The extrusion speed is determined by the die shape information Sk (roundness S
1, k, thickness S2, k, complexity S3, k)
i, k of the antecedent part fitness of i is calculated, and this antecedent part μ
It is calculated by weighted averaging the consequent constant Ui with i and k as weights. Then, the knowledge base inference operation processing 42
Calculates the ideal extrusion rate IV2, k by correcting the preset reference extrusion rate using the calculated extrusion rate.

The ideal extrusion rate IV2, k obtained by the knowledge-based reasoning is the fusion process 23, the knowledge-based reasoning evaluation process.
47 and knowledge data correction processing 48,
It is stored in the input / output history memory 21 and the knowledge base inference output history memory 46.

The knowledge base inference calculation processing 42 may directly calculate the ideal extrusion speed. In this case, the consequent part constant Ui of the rule Ri of the knowledge data is not a relative value with respect to the reference extrusion speed, but an absolute value with respect to the ideal extrusion speed.

2.4.2 Knowledge data learning process

The knowledge base inference output history memory 46 stores the ideal extrusion speed V2,
k is stored for each knowledge base inference in the knowledge base inference operation processing 42.

Since the ideal extrusion speed IV2, k obtained by the knowledge base inference is stored in the input / output history memory 21, the knowledge base inference output history memory 46 is not necessary.

The knowledge base inference evaluation processing 47 is to evaluate the ideal extrusion speed IV2, k obtained by the knowledge base inference to calculate the evaluation value ES2, k.

The evaluation items evaluated by the knowledge base inference evaluation process 47 are knowledge consistency E2,1, k, input space coverage E2,2, k, and past case agreement E2,3, k. is there.

The knowledge consistency E2,1, k is the two rules Ri and Rj having different knowledge data at the present time k.
For (i ≠ j), the antecedent suitability μi, n of the rule Ri for the die shape information Sn (roundness S1, n, thickness S2, n, complexity S3, n) at an arbitrary time point n in the past and Rj μ
Both j and n are high, and the consequent part U of these rules Ri is
That is, i and Uj of Rj do not differ greatly. That is, when two similar rules exist, the consequent constants are almost the same when the antecedent conformity of both rules is high.

The knowledge consistency E2,1, k is the die shape information Sn (roundness S1, n, thickness S2, n, complexity S3) stored in the input / output history memory 21 at any past time n. , n
), Based on the antecedent part suitability μi, n of Ri of the rule Ri of knowledge data stored in the knowledge database 41 and the consequent part constant Ui of Ri, and μj, n of Rj and Uj of Rj, It is calculated by equation (3).

[0199]

[Equation 2]

Where i and j are knowledge database
All rule numbers stored in 41. n is all case numbers stored in the input / output history memory 21.

The coverage E2,2, k of the input space is k at the present time.
For all rules Ri in, the die shape information Sn (roundness S1, n, thickness S2, n,
The antecedent suitability μi of the rule Ri for the complexity S3, n),
This means that there is no one with a small sum of n.

The coverage E2,2, k of the input space is the die shape information Sn (roundness S1, n, thickness S2, n, complexity S3, stored in the input / output history memory 21 at the past time n. n)
For the antecedent part μi, n of each rule Ri calculated using the knowledge data stored in the knowledge data base 41
It is calculated by Eq. (4) based on

[0203]

[Equation 3]

Here, i is all the rule numbers stored in the knowledge database 41. n is all case numbers stored in the input / output history memory 21.

The match E2,3, k of the past cases is k at the present time.
The ideal extrusion rate obtained by performing knowledge base inference on the die shape information Sn (roundness S1, n, thickness S2, n, complexity S3, n) at an arbitrary time point n in the past using the knowledge data in It is the agreement between F (Sn) and the ideal extrusion rate IV2, k at that time point n.

The coincidence E2,3, k of the past cases is determined by the die shape information Sn (roundness S1, n, thickness S2, n, complexity of the past arbitrary time point n stored in the input / output history memory 21. S
3, n), the ideal extrusion speed F (Sn) obtained by performing knowledge base inference using the knowledge data at the present time k stored in the knowledge data base 41, and the knowledge base inference output history memory 46 (or I / O history memory
It is calculated by the equation (5) based on the ideal extrusion speed IV2, n at an arbitrary time point n in the past stored in 21).

[0207]

[Equation 4]

Here, n is all the case numbers stored in the input / output history memory 21.

The knowledge base inference evaluation processing 52 is performed by the equation (3),
Based on the evaluation items E2,1, k, E2,2, k and E2,3, k evaluated by (4) and (5) respectively, the evaluation value ES2, k in the knowledge-based reasoning is given by ).

[0210]

[Equation 5]

In this way, the knowledge base inference evaluation processing is performed.
The evaluation value ES2, k evaluated by 47 is given to the fusion weight calculation processing 22.

The knowledge data modification processing 48 modifies the knowledge data stored in the knowledge data base 41 by learning.

The knowledge data correction processing 48 is performed by the teacher ideal extrusion speed TVn, k and extrusion time T input from the input device 14.
k (manual target extrusion speed pattern VP target extrusion speed V2,
k and speed change time Tk), and quality evaluation information C
Based on k (thickness variation C1, k, sash deformation C2, k),
Learning with meta-knowledge data.

An example of the production rule of meta-knowledge data is shown below.

MR1: If the extrusion time is too long, the consequent part U1 of the rule R1 is made slightly smaller MR2: If the thickness of the aluminum sash is uneven, the consequent part U2 of the rule R2 is made slightly smaller. MR2: If the die shape is flat and the extrusion time is too long, the consequent part U2 of rule R2 and U5 of R5
MR4: If the aluminum sash is deformed, the consequent part U4 of rule R4 and U6 of R6
A little

FIG. 12 shows an example of membership functions of the antecedent part variables and the consequent part variables of the meta-knowledge data.

FIG. 12A is an example of a membership function representing the language information “too long” with respect to the antecedent variable “extrusion time”.

FIG. 12B is an example of a membership function representing the language information “large” with respect to the antecedent variable “wall thickness unevenness”.

FIG. 12C is an example of a membership function representing the language information “large” with respect to the antecedent variable “sash transformation”.

FIG. 12D shows an example of three singletons for the language information “to make it a little smaller” for each rule for the consequent part variable “constant part constant”.

The knowledge data correction processing 48 corrects the knowledge data by performing fuzzy inference according to the meta-knowledge data. In the case of the above meta-knowledge data, only the consequent part of the knowledge data is modified. The membership function of the antecedent variable of the knowledge data may be modified. Alternatively, a new rule may be created and added to the knowledge data.

The learning of the knowledge data uses the teacher ideal extrusion speed TVn, k input from the input device 14 when the operator determines that the ideal extrusion speed IV2, k by the knowledge base inference calculation processing 42 is inappropriate. Knowledge data correction process
The 48 may perform learning by the steepest descent method.

In learning the knowledge data, when the manual target extrusion speed pattern VP is input, the target extrusion speed V2, k of the manual target extrusion speed parameter VP becomes the teacher ideal extrusion speed TVn, k. Ideal extrusion speed TVn, k
Learning is performed using.

As described above, the knowledge base inference is performed, the inference result by the knowledge base inference is evaluated, and the knowledge data is learned.

2.5 Model-based reasoning

The model-based inference processing 50 is to determine the ideal extrusion speed IV3, k by performing model-based inference using a physical model for the die shape information Sk given from the image processing device 13.

The model data learning process 55 evaluates the ideal extrusion rate IV3, k obtained by the model base inference of the model base inference process 50 to calculate the evaluation value ES3, k, and The model data (parameters of the physical model) used by the inference process 50 are modified by learning.

FIG. 13 is a functional block diagram showing the detailed configurations of the model-based inference processing 50 and the model-data learning processing 55. The model-based inference processing 50
Data base 51 and model-based inference processing 52
Consists of. The model data learning process 55 is a model-based inference output history memory 56, a model-based inference evaluation process.
57 and model data correction processing 58. Each of these processes is a program routine.

2.5.1 Model-based inference processing

The model data base 51 stores the model data of the physical model used by the model base inference operation processing 52 for each die. This model data is a parameter of the physical model.

The model-based inference calculation process 52 is performed by the die shape information Sn (roundness S1,
n, thickness S2, n, complexity S3, n)
Using the parameters of the physical model stored in the data base 51, the ideal extrusion rate V3, k is inferred by solving the inverse model of the physical model.

The physical model is composed of the following combinations of sub-models.

Model 1: Physical shape model for determining the shape of molded aluminum sash based on extrusion speed and die shape Model 2: Physical shape of molded aluminum sash, and melting in burette Aluminum formed based on the temperature and heat capacity of the aluminum
Heat distribution model for obtaining heat distribution of sash Model 3: Based on material, physical shape and heat distribution of aluminum sash. Molded aluminum
Thermosetting model for obtaining metallic properties of sash Model 4: Evaluation model for comprehensive evaluation of aluminum sash based on physical shape and metallic properties

The model 1 (physical shape model) is obtained by a simple calculation.

The model 2 (heat distribution model) belongs to the thermodynamic system, and it is impossible to create a complete model considering the influence of disturbance and the like. Model 2 creates a heat distribution model that is approximate to the extent that an inverse model can be created.

The model 3 (thermosetting model) belongs to the chemical system, and like the model 2, a complete physical model cannot be created. Model 3 creates a model approximated to the extent that an inverse model can be created.

The model 4 (evaluation model) is created subjectively.

The model-based inference operation processing 52 solves the inverse model of these physical models to obtain the ideal extrusion rate IV3,
Calculate k. The ideal extrusion rate IV3, k is
It is given to the model-based inference evaluation process 57 and the model data correction process 58, and is also stored in the input / output history memory 21 and the model-based inference output history memory 56.

2.5.2 Model data learning processing

The model-based inference output history memory 56 is
The ideal extrusion speed V3, k given from the model-based inference operation processing 52 is stored for each model-based inference (case number) in the model-based inference operation processing 51.

Since the ideal extrusion speed IV3, k obtained by the model-based inference is stored in the input / output history memory 21, the model-based inference output history memory 56 may be omitted.

The model-based inference evaluation processing 58 evaluates the ideal extrusion rate IV3, k obtained by the model-based inference.

The evaluation item evaluated by the model-based evaluation processing 58 is the coincidence with the past case. The coincidence of the past cases is the evaluation value ES3, k.

The match ES3, k of the past cases is obtained by dice shape information Sn (roundness S1, n at any past time n).
, Thickness S2, n, complexity S3, n), the ideal extrusion speed M (Sn) obtained by model-based inference using the model data at the current time k, and the ideal extrusion speed at that time n. It is consistent with IV3, n.

The match ES3, k of the past cases is the die shape information Sn (roundness S1, n, thickness S2, n, complexity S3, n stored in the input / output history memory 21 at the past time n. )
And the ideal extrusion speed IV3 stored in the model-based inference output history memory 56 (or the input / output history memory 21),
It is calculated by Eq. (7) based on n.

[0246]

[Equation 6]

Here, n is all the case numbers stored in the input / output history memory 21.

The evaluation value ES3, k (calculated according to equation (7)) evaluated by the model-based inference evaluation processing 72 in this way is given to the fusion weight calculation processing 23.

When the operator determines that the ideal extrusion speed V3, k obtained by the model-based inference calculation processing 82 is incorrect, the ideal extrusion speed IVk is input from the input device 14.

The model data correction processing 73 is performed by the model data stored in the model data base 51 (model data
Parameter) is corrected by learning.

The model data correction process 72 is the ideal extrusion speed IV obtained by the model base inference calculation process 52.
3, k and the teacher's ideal extrusion speed T input from the input device 14
Learning by the method of least squares is performed so that there is no error with Vn, k.

When the manual target extrusion speed pattern VPk is input, the manual target extrusion speed V1, k becomes the teacher ideal extrusion speed TVn, k.
The model data is learned using n, k (target extrusion speed V2, k).

As described above, the model-based inference is performed, the inference result by the model-based inference is evaluated, and the model data is learned.

2.6 Fusion Weight Calculation

The fusion weight calculation processing 22 is the evaluation ES1, k by the case-based reasoning evaluation processing 37 and the knowledge-based reasoning evaluation processing.
Based on the evaluation ES2, k by 48 and the model-based inference evaluation processing 58 by the evaluation ES3, k, the ideal extrusion speed IV1, k by the case-based inference processing 30 in the case-based inference processing 30.
, The ideal extrusion rate IV2, k and model-based inference processing by knowledge-based inference in knowledge-based inference processing 40
Ideal extrusion rate IV by model-based reasoning at 50
Fusion weights WT1, k, WT2, k for each of 3, k
And WT3, k (hereinafter, referred to as WTi, k (i = 1, 2 and 3)) are calculated. The fusion weight WTi, k (i = 1, 2 and 3) is fused with the ideal extrusion speed IVi, k (i = 1, 2 and 3) obtained by each inference process in the fusion process 23 described later in detail. It is used for.

14 and 15 are flow charts showing the procedure of the fusion weight calculation processing in the fusion weight calculation processing 22.

I is initialized (FIG. 14; step 71) and the priority adoption flag F1 is set to "FLASE" (FIG. 14).
14; step 72).

The evaluation value ESi, k is the evaluation value ESi, k (i = 1
Is normalized by the maximum value (max ESi, k) of
The evaluation value ESi, k obtained by normalizing the evaluation value ESi, k (ESi, k / max
ESi, k) (FIG. 14; step 73).

The normalized evaluation value ESi, k is the rejection threshold Th0,
i (Fig. 14; step 74). Rejection threshold Th
0, i have different values for each inference method.
The rejection thresholds Th0, i may all have the same value.

The normalized evaluation value ESi, k is the rejection threshold Th0,
If it is determined that it is greater than i (YES in step 74),
The normalized evaluation value ESi, k becomes the fusion weight WTi, k (Fig. 1
4; Step 75).

The normalized evaluation value ESi, k is the rejection threshold Th0,
If it is determined that it is smaller than i (NO in step 74), the fusion weight WTi, k becomes "0" (FIG. 14; step 76).
At this time, since the fusion weight becomes "0", the ideal extrusion speed IVi, k is not used in fusion processing 23 described later (not adopted).

The normalized evaluation value ESi, k is the priority adoption threshold Th.
It is compared with 1, i (Th1, i ≧ Th0, i) (FIG. 14; step 77). The priority adoption threshold Th1, i has a different value for each inference method. The priority adoption threshold Th1, i is
All may have the same value.

The normalized evaluation value ESi, k is the priority adoption threshold Th.
If it is determined to be larger than 1, i (YES in step 77)
), The priority adoption flag F becomes "TRUE" (FIG. 14;
Step 78). In this case, the ideal extrusion rate IVi,
k will be preferentially adopted (priority adoption).

I is incremented (FIG. 14; step 7
9), it is determined whether or not the processes of steps 73 to 80 have been completed for all i (FIG. 14; step 80).

If processing has not been performed for all i (NO in step 80), the process returns to step 73.

When the processing steps 73 to 78 are completed for all i (YES in step 80), it is judged whether or not the priority adoption flag F is "TRUE" (FIG. 1).
4; Step 81).

When the priority adoption flag F becomes "TRUE" in step 78 (YES in step 82),
i is initialized (FIG. 15; step 84).

The normalized evaluation value ESi, k is the priority adoption threshold Th.
It is compared with 1, i (FIG. 15; step 83).

The normalized evaluation value ESi, k is the priority adoption threshold Th.
If it is determined that it is smaller than 1, i (YES in step 83)
), The fusion weight WTi, k becomes "0" (FIG. 15; step 84). That is, the fusion weight WTi, k larger than the priority adoption threshold Th1, i remains as it is, and the other fusion weights become "0".

I is incremented (FIG. 15; step 8
5), it is determined whether or not the processes of steps 83 to 85 have been completed for all i (FIG. 15; step 86).

If processing has not been performed for all i (NO in step 86), the process returns to step 83.

When the processing of the processing steps 83 to 85 has been completed for all i (YES in step 86), the fusion weight calculation processing 22 ends the processing.

If the priority adoption flag F is "FLASE" in step 81 (NO in step 81), the fusion weight calculation process 22 ends.

In this way, the fusion weight calculation processing 22 calculates the fusion weight Wi, k (i = 1 to 3).

The fusion weight WTi, k (i = 1 to 3) is given to the fusion processing 23 and stored in the input / output history memory 21.

2.7 Fusion processing of inference results

The fusion processing 23 is the fusion weight WTi, k (i = 1 to 3) calculated by the fusion weight calculation processing 22, the ideal extrusion speed IV1, k by the case-based reasoning in the case-based reasoning processing 30, and the knowledge base. The ideal extrusion speed IVk is calculated based on the ideal extrusion speed IV2, k by the knowledge-based inference in the inference processing 40 and the ideal extrusion speed IV3, k by the model-based inference processing in the model-based inference processing 50.

The fusion processing 23 is the ideal extrusion speed IV1, k given from the case-based reasoning processing 30 and the knowledge-based reasoning processing.
The ideal extrusion rate IV2, k given from 40 and the model
The ideal extrusion rate IV3, k given from the base inference process 50
For the above, the distance di, j, k between one ideal extrusion speed IVi, k and the other ideal extrusion speed IVj, k is calculated by the equation (8).

[0279]

[Equation 7]

Here, σk is the standard deviation of IVi, k (i = 1 to 3).

For example, ideal extrusion rates IV1, k and IV2,
The distance from k is d1,2, k. The distance di, i, k at the same ideal extrusion speed IVi, k is di, i, k = 1.

The fusion processing 23 is performed by the fusion weight WTi, k (i = 1 to 3) calculated by the fusion weight calculation processing 22 and the ideal extrusion speed IVi, k (i = 1) obtained by each inference processing.
˜3) and their distances di, j, k (i = 1 to 3; j =
1 to 3) and the modified ideal extrusion speed IVi, k (i = 1, 2 and 3) obtained by modifying the ideal extrusion speed obtained by one inference by the other inference according to the equation (10). calculate.

[0283]

[Equation 8]

The fusion processing 23 uses the fusion weights WTi, k (i = 1 to 3) and the distances di, j, k (i = 1 to 3; j = 1 to 3) calculated by the fusion weight calculation processing 22. On the basis of,
By each inference, the reliability CVi, k for the modified ideal extrusion speed Vi, k (i = 1, 2 and 3) is calculated according to the equation (11).

[0285]

[Equation 9]

The fusion processing 22 is performed by the reliability CVi, k (i = 1 to 1).
3) the maximum i is found, and the modified ideal extrusion speed IVi, k with i is determined as the optimum ideal extrusion speed IVk for the die.

This optimum ideal extrusion speed IVk is given to the target extrusion speed pattern creation processing 24 and stored in the input / output history memory 21.

2.8 Creation of target extrusion rate pattern

The target extrusion speed pattern creation processing 24 is the die number Dn, k input from the input device 14 and the fusion processing.
Based on the ideal extrusion rate IVk calculated by 23,
The target extrusion speed pattern is created by standardizing the basic target extrusion speed pattern registered in the basic target extrusion speed pattern memory 25.

The basic target extrusion speed pattern memory 25 is
A target extrusion speed pattern suitable for each die is stored as a basic target extrusion speed pattern. The basic target extrusion speed pattern is not registered for all dies. For dies that are not registered, default basic target extrusion speed patterns are stored.

FIG. 16 shows an example of the default basic target extrusion speed pattern. Similar to the manual target extrusion speed pattern shown in FIG. 3, the basic target extrusion speed pattern is defined by the number of speed changes N, target extrusion speeds V0, V1 and V2, and speed change times T0, T1 and T2.
However, the target extrusion speed V2 is "V2 = 1".

When the die number Dn, k is input from the input device 14, the target extrusion speed pattern creation processing 24 searches the basic target extrusion speed pattern memory 25 and the basic target extrusion speed corresponding to the die number Dn. Read the pattern.

If the basic target extrusion speed pattern for the die number Dn, k is not registered in the basic target extrusion speed pattern memory 25, the target extrusion speed pattern creation processing 24
Reads the default basic target extrusion rate pattern.

When the basic target extrusion speed pattern is read, the target extrusion speed V2 of the basic target extrusion speed pattern is read.
The target extrusion speed pattern is created by normalizing the basic extrusion speed pattern so that the value of the basic extrusion speed matches the ideal extrusion speed IVk given by the fusion process 23.

FIG. 17 shows a target extrusion speed pattern obtained by standardizing the basic target extrusion speed pattern shown in FIG.

The target extrusion speed pattern is given from the target extrusion speed creation processing 24 to the target extrusion speed pattern setting processing 26.

2.9 Setting Target Extrusion Speed Pattern

The target extrusion speed pattern setting process 26 is performed according to "2.
The target extrusion speed pattern created by the target extrusion speed pattern creation process 24 is displayed on the display device (not shown) as described in "1. Outline of target extrusion speed pattern generation". When the operator determines that the target extrusion speed pattern is appropriate, the target extrusion speed pattern setting process 27
Sets the target extrusion speed pattern created by the target extrusion speed pattern creation processing 24 in the control device 11 as a time series of the target extrusion speed. When it is determined to be inappropriate, the manual target extrusion speed pattern input from the input device 14 is set in the control device 11.

As described above, the target value device 10 creates a target extrusion speed pattern and sets it in the control device 11.

[Brief description of drawings]

FIG. 1 is a diagram showing an overall configuration of a control system of an extruder.

FIG. 2 shows an example of a die used for extrusion molding.

FIG. 3 shows an example of a manual extrusion speed pattern.

FIG. 4 shows the evaluation stages of quality evaluation, (A) shows seven evaluation stages for thickness unevenness, and (B) shows seven evaluation stages for sash deformation.

FIG. 5 is a functional block diagram showing a detailed configuration of a target value generation device.

FIG. 6 is a flow chart showing an outline of target extrusion speed pattern generation.

FIG. 7 is a flow chart showing an outline of generation of a target extrusion speed pattern.

FIG. 8 shows a format of data stored in an input / output history memory.

FIG. 9 is a functional block diagram showing a detailed configuration of case-based inference processing and case data learning processing.

FIG. 10 is a functional block diagram showing a detailed configuration of knowledge base inference processing and knowledge data learning processing.

[Fig. 11] Fig. 11 shows an example of a membership function of an antecedent variable and a consequent variable of knowledge data. (A) shows language information "round" and "flat" for the antecedent variable "roundness". (B) shows two membership functions that represent linguistic information “thin” and “thick” regarding the antecedent variable “thickness”, and (C) shows the antecedent variable Linguistic information about "complexity""complex" and "simple"
(D) shows one membership function for each rule with respect to the consequent variable "extrusion speed", and one for "slow down" and 2 for "slightly slow down".
One, two for "make a little faster", and one for "make faster".

[Fig. 12] Fig. 12 shows an example of membership functions of antecedent variables and consequent variables of meta-knowledge data. (A) is a membership function that represents language information "too long" for the antecedent variable "extrusion time". (B) is the antecedent variable “wall thickness unevenness”.
Shows the membership function that represents the linguistic information "large", and (C) shows the membership function that represents the linguistic information "large" with respect to the antecedent variable "sash transformation", and (D).
Shows each single rule for the consequent part variable "constant part constant" and three singletons for the language information "slightly reduce".

FIG. 13 is a functional block diagram illustrating a detailed configuration of model-based inference processing and model-data learning processing.

FIG. 14 is a flow chart showing a processing procedure of fusion weight calculation in fusion weight calculation processing.

FIG. 15 is a flow chart showing a processing procedure of fusion weight calculation in fusion weight calculation processing.

FIG. 16 shows an example of a basic extrusion speed pattern registered in a basic target extrusion speed pattern memory.

FIG. 17 shows an example of a target extrusion speed pattern in which the basic target extrusion speed pattern is standardized. 17 is a target extrusion speed pattern obtained by standardizing the basic target extrusion speed pattern shown in FIG.

[Explanation of symbols]

 1 Extrusion Pump 2 Bullet / Set Machine 3 Die 4 Aluminum Sash 5 Conveyor 6 Tensioner 10 Target Value Generator 11 Controller 12 Camera 13 Image Processor 14 Input Device 21 Input / Output History Memory 22 Fusion Weight Calculation 23 Fusion Fusion 24 Target extrusion speed pattern creation processing 25 Target extrusion speed pattern memory 26 Target extrusion speed pattern setting processing 30 Case-based inference processing 31 Case data base 32 Case-based inference calculation processing 35 Case data learning processing 36 Case-based inference output history memory 37 Case Base inference evaluation processing 38 Case data correction processing 40 Knowledge base inference processing 41 Knowledge data base 42 Knowledge base inference calculation processing 45 Knowledge data learning processing 46 Knowledge base inference output history memory 47 Knowledge base inference evaluation processing 48 Knowledge data correction processing 50 model -Based inference processing 51 Model database 52 Model-based reasoning processing 55 model data learning processing 56 Model-based reasoning output history memory 57 model-based reasoning evaluating process 58 model data correction processing

Front page continuation (72) Inventor Labor Seko 10 Oenron, Hakuen Todocho, Ukyo-ku, Kyoto City, Kyoto Prefecture Omron Co., Ltd. (72) Inventor, Tsutomu Ishida 10th place, Hanazono Tudo-cho, Kyoto City, Kyoto Omron (72) Inventor Yumi Tsutsumi, 10 Hanazono Todocho, Ukyo-ku, Kyoto Prefecture, Kyoto Omron Co., Ltd. (72) Nobuo Nagasaka 10 Hanazono Todocho, Ukyo-ku, Kyoto City, Kyoto Omron Co., Ltd. (72) Inventor Megumi Megumi, Omron Co., Ltd. 10 Hanazono Dodo-cho, Ukyo-ku, Kyoto City, Kyoto Prefecture

Claims (34)

[Claims] 1. A plurality of types of inference means for inferring input information according to separate data bases for controlling a controlled object or processing information, and a plurality of inferences obtained by the plurality of types of inference means. Evaluation means for evaluating each result, output information calculation means for calculating output information based on a plurality of inference results obtained by the plurality of types of inference means, and a plurality of evaluation results obtained by the evaluation means An information processing device comprising: 2. A knowledge base inference means, wherein the plurality of types of inference means make inference according to a knowledge data base,
The information processing apparatus according to claim 1, wherein the information processing apparatus is at least two of a case-based reasoning unit that performs reasoning according to a case data base and a model-based reasoning unit that performs reasoning according to a model data base. 3. The fusion weight calculation in which the output information calculation means calculates fusion weights by normalizing each of the plurality of evaluation results obtained by the evaluation means with the maximum value of the evaluation results. And inference result fusion means for calculating output information based on the plurality of inference results obtained by the plurality of types of inference means and the plurality of fusion weights calculated by the fusion weight calculation means. The information processing device according to claim 1. 4. The information processing apparatus according to claim 3, wherein the fusion weight calculation means sets the fusion weight to 0 when the calculated fusion weight is smaller than a predetermined rejection threshold. 5. The fusion weight calculation means sets a fusion weight other than the fusion weight to 0 when there is a fusion weight equal to or higher than a predetermined priority adoption threshold among the calculated fusion weights. The information processing device according to item 3. 6. The inference result merging means includes a plurality of inference results obtained by the plurality of types of inference means.
The distance between the one inference result and the other inference result is calculated, and the modified inference result obtained by modifying the one inference result is calculated based on the other inference result and the distance. The output information is calculated by calculating the reliability of each of the plurality of modified inference results based on the calculated plurality of fusion weights and the plurality of distances, and selecting the modified inference result that maximizes the reliability. The information processing apparatus according to claim 3, which is for calculating. 7. The information processing apparatus according to claim 1, further comprising learning means for correcting each of the separate data bases by learning based on inputted teacher information. 8. A plurality of types of inference are performed on input information according to separate data bases for controlling a controlled object or processing information, and a plurality of inference results obtained by the plurality of types of inference are evaluated respectively. And an output information is calculated based on a plurality of inference results obtained by the plurality of types of inference and a plurality of evaluation results obtained by the evaluation. 9. A model in which the plurality of types of inferences make inferences based on a knowledge database, a knowledge base inference that makes inferences based on a case data base, and a model that makes inferences based on a model data base. The information processing method according to claim 8, which is at least any two of the base inferences. 10. The plurality of evaluation results obtained by the evaluation are normalized by the maximum value of the evaluation results to calculate fusion weights, respectively, and the plurality of results obtained by the plurality of types of inference are calculated. The information processing method according to claim 8, wherein the output information is calculated based on the inference result and the plurality of calculated fusion weights. 11. The fusion weight is set to 0 when the calculated fusion weight is smaller than a predetermined rejection threshold.
Information processing method described in. 12. The information processing method according to claim 10, wherein when there is a fusion weight that is equal to or greater than a predetermined priority adoption threshold value among the calculated fusion weights, fusion weights other than the fusion weight are set to 0. 13. With respect to a plurality of inference results obtained by the plurality of types of inference, distances between one inference result and another inference result are respectively calculated, and the distance is calculated based on the other inference result and the distance. A modified inference result obtained by modifying one inference result is calculated, and reliability is calculated for each of the plurality of modified inference results based on the plurality of fusion weights and the plurality of distances. 11. The information processing method according to claim 10, wherein the output information is calculated by selecting the modified inference result. 14. The information processing method according to claim 8, wherein each of the separate data bases is modified by learning based on inputted teacher information. 15. A first sensor for measuring a controlled variable of a controlled object, an actuator for operating the controlled object based on a given manipulated variable, and a controlled variable measured by the first sensor for a given target. Determine the operation amount so that it approaches the amount,
In a control system including a control device for giving to the actuator, a target amount generation device that sets a target amount in the control device is measured by a second sensor that measures target information regarding the control target and the second sensor. The target information and the input information to be input are respectively inferred according to separate data bases for generating the target amount, and a plurality of types of inference means for calculating the target amount are respectively obtained by the plurality of types of inference means. By evaluating each of the obtained plurality of target quantities and calculating the respective evaluation values, by normalizing each of the plurality of evaluation values obtained by the above evaluation means by the maximum value of the evaluation values, Fusion weight calculation means for calculating fusion weights respectively, and a plurality of fusion weights obtained by the inference means Target amount and the target amount fusion means for calculating an optimum target amount for the controlled object based on a plurality of fusion weight calculated by the fusion weight calculation means,
The target quantity is set by standardizing the preset basic target quantity pattern for setting the control quantity to the target quantity quickly and preventing the overshoot quantity from occurring by using the optimum target quantity calculated by the target quantity fusion means. A target quantity pattern creating means for creating a pattern, and a learning means for modifying each of the separate data bases by learning based on input teacher information, and the target quantity pattern created by the target quantity pattern creating means A control system for providing the above-mentioned control device. 16. A first sensor for measuring a controlled variable of a controlled object, an actuator for operating the controlled object based on a given manipulated variable, and a controlled variable measured by the first sensor for a given target. Determine the operation amount so that it approaches the amount,
In a control system including a control device for giving to the actuator, a target amount generation device that sets a target amount in the control device is measured by a second sensor that measures target information regarding the control target and the second sensor. The target information and the input information to be input are inferred according to separate data bases for generating the target amount, and a plurality of types of inference means for calculating the target amount are respectively obtained. Evaluating means for evaluating each of the obtained plurality of target amounts and calculating respective evaluation values, a plurality of target amounts obtained by the plurality of types of inference means, and a plurality of evaluation values obtained by the above-mentioned evaluating means A target amount fusion means for calculating an optimum target amount for the controlled object based on
A control system for giving the above-mentioned control device the optimum target amount calculated by the above-mentioned target amount fusion means. 17. A knowledge base inference means, wherein the plurality of types of inference means make inference according to a knowledge data base,
At least 2 of case-based reasoning means for reasoning according to a case data base and model-based reasoning means for reasoning according to a model data base
17. The control system according to any one of claims 15 or 16 which is one. 18. A fusion weight calculation in which the target amount calculation means calculates fusion weights by normalizing each of the plurality of evaluation values obtained by the evaluation means with the maximum value of the evaluation values. Means, and a plurality of target quantities obtained by the plurality of types of inference means,
17. The control system according to claim 16, further comprising target amount fusion means for calculating the optimum target amount based on a plurality of fusion weights calculated by the fusion weight calculation means. 19. The control according to claim 15, wherein the fusion weight calculation means sets the fusion weight to 0 when the calculated fusion weight is smaller than a predetermined rejection threshold. system. 20. The fusion weight calculation means sets a fusion weight other than the fusion weight to 0 when there is a fusion weight equal to or higher than a predetermined priority adoption threshold among the calculated fusion weights. The control system according to any one of 15 or 18. 21. The target amount fusion means calculates a distance between one target amount and another target amount for each of the plurality of target amounts obtained by the plurality of types of inference means, and the other target amount is calculated. And the distances thereof to calculate modified target amounts by modifying the one target amount, and based on the plurality of fusion weights and the plurality of distances calculated by the fusion weight means, the plurality of modified target amounts. 19. The optimal target amount is calculated by calculating the reliability of each of the above, and selecting the modified target amount that maximizes the reliability. 19. Control system. 22. The control system according to claim 16, wherein the target amount generation device further comprises learning means for correcting each of the separate data bases by learning based on input teacher information. 23. The optimum amount calculated by the target amount calculation means by the target amount generation device for a preset basic target amount pattern for the control amount to quickly settle to the target amount and to prevent an overshoot amount from occurring. 17. A target amount pattern creating means for creating a target amount pattern by normalizing using a different target amount, the target amount pattern created by the target amount pattern creating means is given to the control device. The control system described in. 24. The control system according to claim 15, which controls an extrusion speed in an extruder. 25. A first sensor for measuring a controlled variable of a controlled object, an actuator for manipulating the controlled object based on a given manipulated variable, and a controlled variable measured by the first sensor for a given target. Determine the operation amount so that it approaches the amount,
In a control system including a control device for giving to the actuator, object information regarding the controlled object is measured, and a separate data base for generating a target amount is measured for the measured object information and input input information. A plurality of types of inference are performed, target amounts are calculated, the above-mentioned separate data bases are modified by learning based on the input teacher information, and a plurality of target amounts obtained by the above-mentioned plurality of types of inference are calculated. An evaluation is performed, each evaluation value is calculated, and the fusion weights are calculated by normalizing each evaluation value with respect to the maximum evaluation value for the plurality of evaluation values obtained by the above evaluation. Based on a plurality of target quantities obtained by inference and a plurality of calculated fusion weights, To calculate the optimum target amount, and to standardize the preset basic target amount pattern that sets the control amount to the target amount quickly and does not cause overshoot by using the calculated optimum target amount. A control method in which a target amount pattern is created by, and the created target amount pattern is given to the control device. 26. A first sensor for measuring a controlled variable of a controlled object, an actuator for manipulating the controlled object based on a given manipulated variable, and a controlled variable measured by the first sensor for a given target. Determine the operation amount so that it approaches the amount,
In a control system including a control device for giving to the actuator, object information regarding the controlled object is measured, and a separate data base for generating a target amount is measured for the measured object information and input input information. Multiple types of inference are performed, target amounts are respectively calculated, multiple target amounts obtained by the multiple types of inference are respectively evaluated, evaluation values are calculated, and multiple types obtained by the multiple types of inference are calculated. And a plurality of evaluation values obtained by the above-mentioned evaluation, an optimum target amount for the above-mentioned controlled object is calculated, and the calculated optimum target amount is given to the above-mentioned control device. 27. A model in which the plurality of types of inferences make inferences based on a knowledge database, a knowledge base inference that makes inferences based on a case data base, and a model that makes inferences based on a model data base. The claim 25 being at least two of the base inferences
Alternatively, the control method according to any one of 26. 28. With respect to a plurality of evaluation values obtained by the evaluation, the fusion weights are respectively calculated by normalizing the respective evaluation values by the maximum value of the evaluation values, and the plurality of inference values obtained by the plurality of types of inference are calculated. 27. The control method according to claim 26, wherein the optimum target amount is calculated on the basis of the target amount and the plurality of calculated fusion weights. 29. When the calculated fusion weight is smaller than a predetermined rejection threshold, the fusion weight is set to 0.
Or the control method according to any one of 28. 30. The fusion weight other than the fusion weight is set to 0 when there is a fusion weight equal to or more than a predetermined priority adoption threshold among the calculated fusion weights. The described control method. 31. With respect to a plurality of target amounts obtained by the plurality of types of inference, distances between one target amount and another target amount are respectively calculated, and based on the other target amounts and the distances, A correction target amount obtained by correcting one target amount is calculated, and the reliability of each of the plurality of correction target amounts is calculated based on the plurality of fusion weights and the plurality of distances. 29. The control method according to claim 25, wherein the optimum target amount is calculated by selecting the corrected target amount. 32. The control method according to claim 26, wherein each of the separate databases is modified by learning based on inputted teacher information. 33. A target quantity pattern is obtained by normalizing a preset basic target quantity pattern for setting the control quantity to the target quantity quickly and preventing an overshoot from occurring by using the calculated optimum target quantity. 27. The control method according to claim 26, wherein the control device is provided with the created target amount pattern. 34. The control method according to any one of claims 25 to 33, which controls an extrusion speed in an extruder.