JP3403432B2 - Machining result prediction device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses a rule base.
From the processing condition data or fixed condition data
Is a device that predicts
The present invention relates to a device that can be chemically optimized. 2. Description of the Related Art Conventionally, processing condition data such as a feed speed and a grinding wheel have been known.
Input conditions consisting of fixed conditions such as the grain size of the workpiece and the finished diameter of the workpiece
Case data is obtained by skilled personnel in accordance with the processing requirements of the workpiece.
It is required empirically. Therefore, the input condition data
Predict the processing result when processing a workpiece with a table
It has become essential. [0003] From such input condition data, processing result data
As a method to predict data, input condition data and processing results
Build a rule base with causal relationships with data
And apply the rules in the rule base to predict the machining results.
There is a method of measuring. This rule-based rule
Is the antecedent part that defines the preconditions for the rule
It consists of a consequent part that stipulates the conclusion of the conclusion. Immediately
That is, the rules are defined by so-called if and then rules.
You. Such rules are empirical knowledge, expert intuition, rules
According to the above, a large number are created as a rule base. [0004] However, such
Creating a rule base is very difficult. why
Then, a certain conclusion for a certain precondition
Although it is easy to correspond, multiple conditions are organic
If there is a correlation, the whole conclusion is that
It is not just the sum of the conclusions corresponding to each condition. Follow
To create rules that have such interrelationships
Is quite difficult. [0005] Also, the rule base is used for all events from the beginning.
May not always be desirable,
In the process of use, unnecessary rules and rules to add
appear. In such cases, rule-based maintenance is
It is performed, but the improperity that has occurred in actual use
Remember the examples and rebuild the rule base regularly
Had to be done. [0006] In addition, the rebuilding of such a rule base is as follows:
Inapplicability caused by past practical use by skilled persons
This has been done on an example basis. And the self of the optimal rule
Devices that automatically generate and change the self-generation of rules are not known.
Not been. In other words, the rule base is optimally self-evolving
There is no device to convert. The present invention has been made to solve the above problems.
The purpose is to optimize the prediction of machining results
Rule-based self-evolutionary used to predict machining results
Is to perform optimal optimization. This rule-based optimization
Between the input condition data and the machining result data
Optimal causal relationships to optimize the prediction of machining results
It is to be. Means for Solving the Problems [0008] The structure of the present invention for solving the above-mentioned problems.
As shown in FIG. 13, as shown in FIG.
Rank in the range of values that can be processed with the result as the consequent part
Rule base memorized as multiple rules with divided values
Storage means X1 and input condition data for processing a workpiece
Input condition data input means X2 for inputting data,
Apply the antecedent part of each rule to the data and match
Processing results to obtain processing result data based on the consequent part of the rule
Result data calculation means X3 and the evaluation value of the processing result data.
And, according to the accumulation of the evaluation value, the conformity of the rule
A fitness calculating means X4 for calculating the fitness that represents
Level storage that stores the level of conformity of each rule in association with each rule
According to the means X5 and the degree of conformity of each rule, the degree of conformity is low.
Sorting method that deletes rules that do not match and leaves rules with high relevance
X6, and among the rules, the conformity is larger than a predetermined value,
Also, select any two rules that have the same consequent part, and
Condition elements that constitute the antecedent part between the two rules
By exchanging any number of corresponding condition elements
Rule generation means X7 for generating two new rules
That is the provision. First, the factor between input condition data and processing result data
A large number of result relationships are prepared as rules. The rule is i
It is described in f, then format. And the antecedent part (if part)
Is the condition part that describes the preconditions for the input condition data.
Minutes, and the consequent part (then part)
This is the conclusion that describes the processing result data when the
It is possible to perform the processing confirmed experimentally for each
Are divided into multiple ranks (for example, from 0.1 to 0.2
Is rank 1). [0010] Depending on the required finishing value,
Processing condition data and fixed conditions predetermined by intuition
Input condition data consisting of data and input condition data input
Input by means. And the processing result data operator
By the stage, the rule based on the input condition data
Rules apply. That is, the antecedent part of the rule is
All rules that match the case data are extracted. And
Processing result data is calculated based on the consequent part of this rule.
It is. This processing result data must be
Judging the result of actually processing a workpiece with data
Is evaluated by [0011] Such a rule base is applicable to various processing.
Is actually used for input condition data
Work result data is obtained. And in the process, a skilled person
Evaluation of each processing result data
Is done. For example, in the case of such a set of input condition data
In addition, the set of such processing result data obtained is "true"
Or "false". Immediately
That is, the application rule is evaluated. like this
Evaluations are performed on various real-world events. Soshi
Therefore, by the fitness calculation means,
Thus, the degree of conformity indicating the conformity of each rule is calculated. Immediately
That is, the degree of conformity of each rule is
It shows how relevant it is. Each of these
The fitness for each file is stored in the fitness calculating means. Thus, the rule base is used for a predetermined period.
After that, when the rule base is updated, the sorting means
The rule is selected according to the conformity of the rule,
Rules in real use can lead to certain conclusions
It means that it is used to a lesser degree, so the rules
Deleted from base. Also, from the rule base,
Any degree that the degree of integration is greater than the predetermined value and the consequent part is equal
Two data are selected. That is, to reach the same conclusion
Any two rules that have contributed a lot are selected
You. Then, the two data are output by the rule generation means.
Any number of pairs in the multiple conditional elements that constitute the antecedent part of
The corresponding condition elements can be exchanged with each other.
Generates two new rules. Two rules with high relevance leading to the same conclusion
By exchanging between corresponding conditional elements in the antecedent part in
Highly similar to the previous two rules, and both rules
New rules that combine the properties of
Is done. Good enough for such rule generation
High degree and the two previous benign components (causal relationship with conclusions)
Benign rules inheriting each other
Could be Benign le inheriting this benign component
Rules are used to determine how well the rules
Is higher and rule-based updates
Survive in the process. On the other hand, two rules are inferior
Inferiority inherited (conditional element with weak causal relation to conclusion)
Rules are also generated. But the rule of inferiority is that
Later use will result in poor fit and eventually rule-based
Is destined to disappear from the country. The previous two rules also
Is left in the rule base, and the rule will survive in the future.
Whether or not to follow that rule in real life use
Is determined by the fitness of In this manner, the rule base is actually used.
And by updating it regularly,
The rules that inherited the rules will survive,
Ultimately, the rule base is
It will be optimized. This optimized loop
If the base is used, the optimal
Obviously, you can obtain composite processing result data,
Is applied in reverse, that is, certain processing result data and the consequent part match
Corresponding input condition data based on the antecedent part of the rule
Data can be obtained. That is, the data with the processing result data
If data items are modified, the
It is possible to obtain appropriate input condition data. According to the present invention, the rule base storage means, the input condition
Corresponding to the input condition data input by the
And apply the antecedent part of each rule, and
Processing result data to obtain processing result data based on the part
Calculation means, so that
It is possible to predict the processing result when the work is performed.
Also, the present invention provides a method of calculating fitness, selecting means, and generating rules.
By providing the means, each evaluation from actual use
The conformity of the rule is determined, and the rule is set according to the conformity.
Are alive or annihilated and any two
Benign departments with strong causal relationship to conclusions
Benign rules that inherit from are generated in a self-proliferating manner. So
And its benign rules are highly compatible with subsequent use
In the rule-based generational shift
survive. As a result, according to the processing result prediction device of the present invention,
In the end, the rule-base is ultimately
A set of rules with complex conditional elements
That is, rule-based optimization
You. In addition, values within the range that can be processed are divided into multiple ranks.
Values are stored as rules, so feed rates etc.
Rules can be created using only the quantitative values of
Can be easily performed. Hereinafter, the present invention will be described based on specific examples.
I will tell. In this embodiment, the processing conditions and the fixed conditions
Used to predict the results and the processing results
Prediction equipment with a self-evolving rule base
Is used for a grinding machine. (1) Configuration of Grinding Machine FIG. 1 has a rule-based optimization device according to the present invention.
FIG. 2 is a configuration diagram showing an overall mechanical configuration of the numerically controlled grinding machine.
You. 50 is a grinder, and a bed 51 of the grinder 50 is provided.
On the table, a table 5 that slides against the bed 51
2 are provided. The table 52 is a table feed mode.
Move in the horizontal direction of the drawing by driving the motor 53
Is done. A headstock 54 and a tailstock are placed on the table 52.
A headstock 54 has a spindle 55,
The tailstock 56 has a tailstock shaft 57. The workpiece W is moved by a spindle 55 and a tailstock 57.
And is rotated by rotation of the main shaft 55. this
The rotation of the spindle 55 is performed by the spindle motor 5 disposed on the headstock 54.
9. On the other hand, a grinding wheel for grinding the workpiece W
Reference numeral 60 denotes a grinding wheel drive motor 62 provided on a grinding wheel stand 61.
It is supported by the drive shaft. Also, the grinding wheel head 61 feeds the grinding wheel head
The motor 63 is controlled to move in the vertical direction in the drawing.
You. Table feed motor 53, wheel head feed motor 6
3. Drives spindle motor 59, grinding wheel drive motor 62, etc.
A numerical controller 30 is provided for control. (2) Configuration of Numerical Control Unit Numerical control unit 30 is mainly composed of C as shown in FIG.
PU31, ROM32, RAM33 and IF (interface
Source 34). NC in RAM33
NC data area 331 for storing programs and described later
Input conditions for storing machining condition data and fixed condition data
Stores data area 332 and predicted machining result data
Processing result data storage area 333 and the rule base
Rule base area 334 to be stored and the degree of conformity of each rule
Is stored. The RAM 33 is backed up by a battery.
Have been. Also, it is stored in the rule base area 334.
Rule bases and rules stored in the fitness area 335
The degree of conformity of the tool is fixed via the interface 34.
Disk device 35 and the fixed disk device 35
And store those data in those areas of the RAM 33.
Can be made. The ROM 32 has an NC data area 33.
The machining control is performed according to the NC data stored in 1.
Control program area storing control programs for
321 and from the input condition data by applying the rule base
A machining result prediction program that predicts machining result data is recorded.
The machining result prediction program area 322
Input condition data when a part of the processing result data is changed
An input condition that stores the desired input condition data calculation program
Optimize the condition calculation program area 323 and rule base
Rule-based optimization program that stores
A program area 324 is provided. (3) Types of data Machining condition data is used to machine a workpiece to a desired finished state.
Data indicating the operating state of the grinding machine for performing the grinding. Real truth
In the embodiment, the processing condition data is, as shown in FIG.
There are 11 items from the rough grinding rotation speed to the stop time after fine grinding. The fixed condition data is used for finishing a certain workpiece to a desired finish.
Material, grinding wheel material and finish for machining
With fixed data that cannot be changed in response to requirements such as
is there. In this embodiment, the fixed condition data is shown in FIG.
As described above, there are nine items, that is, the material of the abrasive grain and the presence or absence of the fixed size. this
Input the processing condition data and fixed condition data
It is called condition data. The processing result data is used when the workpiece is processed.
This is data indicating a raised state. In this embodiment,
As shown in FIG. 3, the result data has eight terms of surface roughness to runout.
Eyes. (4) Rule Configuration Rules describe the relationship between input condition data and processing result data.
It is described. The rules are "if, if"
The if part that describes the matter and the th that describes the conclusion of
It is described by the so-called if, then rule consisting of the en part.
ing. Hereinafter, the if section is referred to as the antecedent section, and the then section is referred to as the consequent section. The rule data is structured as shown in FIG.
Have been. Input condition data and processing result data
Data items are assigned in the order shown. That is,
Data items and data positions are fixed and correspond
You. Each data item has an equal fixed bit length.
You. Therefore, in the case of the present embodiment, the rule is 28 data.
Data items, the bit length of each data item must be 4
The bit length of the rule is 112 bits.
You. Each data item is composed of 4 bits.
Therefore, if there is no value set for that data item,
As "0000", the data of one item is "0001" (1) to "111".
A maximum of 15 rank values of 1 "(15) can be displayed.
FIG. 5 shows how data is ranked. What
In this ranking, grinding is actually performed in each data.
Experimentally find possible data values and divide them equally
Have been. Thus, the value of each data item in the rule
Is given by the rank value of the data. Antecedent and after
Each data item in the subject part is connected by a logical operation and
Have been. "#" When no value is set for the data item
Is displayed with. Hereinafter, in general, the rules are as shown in FIG.
write. n is the number of rules in the database, a
_i (j) is the i-th rule R _i Antecedent part (input condition data
Part) represents the value (rank value) of the j-th data item.
And b _i (j) is the i-th rule R _i Consequent part (processing result
Data part) in the j-th data item (rank value)
Represent. p and q are the data terms of the antecedent part and the consequent part, respectively.
The number of eyes. (5) Creation of Initial Rule The initial rule is created as shown in FIG. That is,
An empirical rule that chatter occurs at high feed rates
If there is, the rough grinding feed speed rank value and chatter
Empirically, the relationship between the feed rate and the rough
Created for link value (15). As a result, 15
A rule is created. For example, one rule is "
If the rank value of the vibration speed is 5, the rank value of the chatter is 1
There is ". This rule is represented by a structure shown in FIG.
And the seventh data item in the antecedent part of the rule is as shown in FIG.
It is "Rough feed rate", and the fourth data item in the consequent
Since the eyes are "chattering", _t な る [###### 5 #############] [### 1 ####] Similarly, the relationship between the fine feed rate and chatter is also required.
And five rules are created. In the same way, "the chatter where the rigidity of the workpiece is low
Occurs, "the stiffness coefficient
A correspondence between the rank value and the chat rank value is required,
Five rules corresponding to all ranks (5) of rigidity coefficient
Created. For example, one rule is "run stiffness
If the peak value is 2, the chatter rank value is 2. "
You. This rule is represented by a structure shown in FIG.
As shown in FIG. 4, the fifteenth data item of the antecedent part is
"Workpiece rigidity coefficient" is the fourth data item in the consequent part
Is “chatter”, so that _w な る [############## 2 #####] [### 2 ####] Hereinafter, similarly, the rules of the structure shown in FIG.
Created and its rules are stored on a fixed disk with a rule base.
Is memorized. (6) Input condition given when processing a workpiece
When processing a workpiece with relevant data, consider various finishing requirements.
In consideration of this, the operator must enter values for all items in the input condition data.
To determine. This input condition data is
Alternatively, it is obtained by calculation by a computer.
This value is not a rank value but an absolute value. For example, coarse
The rotation speed is given as 300 rpm. Below, general
Then, the input condition data given to this reality is [D (1) ~
D (p)]. That is, D (i) is the value of the i-th data item.
(Rank value). It is necessary to indicate each data item
When there is no input condition data, the input condition data
Data D. (7) Application of Rules Next, when the actual input condition data D is given,
Before processing the crops in real time, predict the processing results.
And actually add a workpiece using the input condition data D.
Work is done. In the former case, the input condition
A processing result is predicted using the data D in the following procedure.
Also, in the latter case, the processing result is
When a request was made to improve the results a little,
The requested value is entered. Then, the correction
Force condition data is required. In this case also,
Before obtaining the data, the first input condition data D
Corresponding to the input condition data D by the following procedure using
The processing result to be predicted is performed. FIG.
Stored in the M32 machining result prediction program area 322
Result prediction processing executed by the CPU 31
It is a flowchart of a gram. In step 100, the rule base and the
And the conformity of each rule are input from the fixed disk 35,
Rule base area 334 and suitability area 3 of RAM 33
35. Next, in step 102, the key
The actual input condition data D is input from the board 22,
It is stored in the input condition data area 332 of the RAM 33.
Next, in step 104, the input condition data D
Is converted into a rank value by the ranking shown in FIG.
And the input condition data [D (1) to D
(p)] is stored in the input condition data area 332. Next, in step 106, the data item
The eye number j is initially set to 1. Next, step 108
, All of the rules stored in the rule base area 334
The rule is applied to the input condition data D. That is,
First data item of each of input condition data and rule
D (1) = a _i All rules that satisfy (1)
Is extracted. Next, in step 110, the input condition
Rule application completed for all data items of the case data
It is determined whether or not the operation has been completed.
The processing of step 108 is repeated. As a result,
Data and the value of any j-th data item of the rule
And D (j) = a _i All rules that satisfy (j) are extracted
You. When the rule extraction is completed, step 11
In step 2, after all the rules extracted in step 108
In the data section, for each data item,
Is determined. The set of values is
, B (q)]. still,
When the processing result data is collectively referred to as processing result data B
Describe. That is, the actual input condition data created by the operator
If the workpiece is machined in D, finish the workpiece
When the state shown in machining result data B is reached,
It is determined by the rule base of the stage. Next, at step 114, step
All rules extracted in step 108 and processing result data B
Rules with equal values are extracted for any data item
Is done. That is, for each data item, B (j) = b _i (j)
The rule that satisfies is extracted. This extracted rule is
Generate the value B (j) of the j-th data item of the processing result data
Rules. Next, at step 116, the operator
Work with experience or with actual input condition data D
Judging from the finished state of the workpiece obtained by processing the workpiece
Evaluation is given to each data item of the processing result data B.
available. This evaluation is based on each data item of the processing result data B.
Passed “1” if the eye value is valid, unacceptable if not valid
The case is binaryly determined to be “0”. This judgment result is input
It is. Next, at step 118, a pass judgment is made.
The value B (m) of the data item of the machining result data
Rule R created _k Of conformity F _k Is updated. still,
The initial value of the matching degree of all rules is 0. This rule
Is updated as follows. Now update the conformity
The new rule is [Equation 3] Rule R _k ｛[############## 2 #####] [# 1 # 2 ### 3]｝, and in the consequent part, data obtained as a pass
Assume that the items were a second data item and an eighth data item
You. At this time, in the consequent part,
The number of data items 2 is the current fitness F _k Is added to (F
_k = F _k +2). That is, a pass judgment is obtained for a plurality of data items.
The rules that have been applied to the input condition data D
It is determined that the rule is high. In addition, data of pass judgment
Rules with no items have low relevance, so
No numerical updates are made. Next, in step 120
The rule base stored in the rule base area 335.
Source is output to a fixed disk together with the degree of conformity of the rule.
It is memorized. (8) Correction of input condition data Work on the processing result data B obtained by the program of (7)
Changes the value of a data item,
Input condition data to obtain processed data
There are times when it does. For example, obtained processing result data
When the rank value of the surface roughness of B is "7", the rank value
Improve the surface roughness by "3" and work with rank value "4"
A request to finish is generated. At this time, the surface roughness
Input condition data to achieve rank value "4"
Need to be Also, using the input condition data D
The operator judges the result of machining and changes a certain machining result.
Sometimes you want to be good. For example, the current state of surface roughness
The rank value is not known exactly, but the surface roughness is
Wants to improve the rank value "3" relatively
There is. At this time, run the surface roughness more than the current rank value.
Corrected input condition data necessary to improve only by "3"
Need to ask. In this case, the CPU 31
FIG. 9 stored in the input condition calculation program area of FIG.
Execute the input condition calculation program shown. Step 20
At 0, the rule base and the conformity of each rule are fixed
From the disk device 35 to the rule base area 3 of the RAM 33
34 and read into the fitness area 335. Next,
In step 202, the processing result data is
In Data B, the value of the data item
You. The processing result data after the change is [B ^' (1),…, B
^' (q)]. The generic name of the processing result data after this change is
Processing result data B after change ^' Is described. This processing result
To change data, specify the data item you want to change and
The operator enters the rank value H of the relative change amount of the data item.
It is done by force. And the processing result data B
And processing result data after change using relative rank value H
B ^' Is calculated. Next, in step 204, the data item
The eye number j is initially set to 1. Next, step 206
, All of the rules stored in the rule base area 334
The rule is inversely applied to the processing result data B. Immediately
First, the modified processing result data and the first data of the consequent part of the rule
For the value of the data item, B ^' (1) = b _i Everything that satisfies (1)
Are extracted. Next, in step 208
And processing result data B after change ^' Related to all data items
To determine whether the rule application has been completed.
If not, the process of step 206 is repeated.
You. Thereby, the post-change processing result data B ^' And the rules
Regarding the value of any j-th data item in the consequent, B ^' (j)
= B _i All rules that satisfy (j) are extracted. When the rule extraction is completed, step 21
0, before all rules extracted in step 206
In the data section, for each data item,
Is determined. The set of values is
Force condition data [A (1),..., A (q)]. Typically
Is referred to as modified input condition data A. That is, after the change
Processing result data B ^' Input condition data A for obtaining
Is required. Next, in step 212, step
All the rules extracted in 206 and the modified input condition data A
For any data item in
Is extracted. That is, for each data item, A (j) = a
_i Rules that satisfy (j) are extracted. This extracted le
Is the value A of the j-th data item of the corrected input condition data A.
This is the rule that created (j). The modified input condition data [A (1),.
(q)] is not always determined for all data items
Absent. That is, any of the routes extracted in step 206
Data items for which no rank value is described even in the antecedent part of
Can exist. In this case, this data of the input condition data
Data items will not be determined by the rules
And for that data item, the first input condition data
The value of the corresponding data item of D is used as it is. Next, at step 214, the operator's experience or
Is obtained by actually processing a workpiece with the corrected input condition data A.
Judging from the finished state of the workpiece to be
An evaluation is given to each data item of data A. This
Is evaluated when the value of each data item of the corrected input condition data A is
Pass "1" if valid, fail "0" otherwise
Is determined in a binary manner. This determination result is input. Next, at step 216, a pass judgment is made.
The rule that produced the value A (m) of the data item for which the
R _k Of conformity F _k Is updated. This rule ’s conformance
The update is performed as follows. Now, update the fitness
Le is [Equation 4] Rule R _k ｛[# 2 ## 4 ## 5 ###### 2 #####] [4 #######]｝, and the data obtained as a pass in the antecedent
It is assumed that the items are the second, fifth, and fifteenth data items.
At this time, in the antecedent part, data obtained as a pass
The number 3 of items is the current fitness F _k Is added to (F _k =
F _k +3). That is, a pass judgment is obtained for a plurality of data items.
Rule is determined to be a rule with a higher degree of conformity.
You. Note that rules that do not have a data item for pass judgment are appropriate.
Since the fitness is low, the value of the fitness is not updated. Next
In step 214, the rule base area 335
The rule base stored in
Output to a fixed disk and stored. (9) Prediction of Effect of Processing Result Next, the input condition data [D (1),..., D (q)] shown in FIG.
Instead, the above modified input condition data [A (1), ..., A
(q)] to create a rule base according to the procedure shown in FIG.
By applying, the modified input condition data [A (1),
…, A (q)] when processing the workpiece
Fruit data [B ^" (1),…, B ^" (q)]
You. That is, a data item of the processing result data, for example,
Fix to change the surface roughness rank value according to the required value
Influence on other processing results when processing with positive input condition data A
Can be predicted. This processing result data
Data B ^" , And make changes if necessary
You can also. (10) Rule Base Learning Next, the rule base is executed when it becomes necessary to update the rule base.
The following describes a rule-based optimization procedure to be performed. FIG.
0 is the rule of ROM32 executed by CPU31
The rules stored in the base optimization program area 324
5 is a flowchart of a rule-based optimization program.
In step 300, the data is stored in the fixed disk device 35.
The rule base and the conformity of each rule are stored in RAM3.
3 in the rule base area 334 and the fitness level area 335
Be included. Next, in step 302, all rules
Of conformity F _k Considering the maximum value of
Value T _H, Lower threshold T _L Is determined. Then, step
At 304, the fitness F _k Is the threshold T _H More roux
Is extracted. Then, for each of the extracted rules,
Then, the conformity F of each rule _k According to the number of identical
A rule is generated. The conformity of the generated rule is
It has the same value as the conformance of the original rule. Next, in step 306, the degree of conformity is
Threshold T _L The following rules are extinguished. Next,
In step 310, the degree of conformity is equal to the threshold T. _L And threshold T
_H The rules between and remain. More than
Steps leave rules with high relevance and low relevance
New rules are extinguished. In other words, rules with low conformance are
Since these rules do not contribute to the correct result,
To be extinguished from the base. For it
Therefore, rules with high relevance are rules that give reasonable results.
Therefore, it is saved in the rule-based update.
Also, multiple identical rules with high relevance are generated.
Data cross-over performed in the following steps
Rules generated by the server
Rules related to many rules are easily generated. Follow
Time to the optimal rule base
Become. Next, in step 312, all rules
Of which are greater than a predetermined threshold and
Rules where the value of each data item is equal
All rules are grouped for each value of the consequent. For example,
The rule group in which the value of the first data item of the consequent is “4” is
As shown in FIG. Any two from this rule group
Are extracted. For example, Rule No.1 and Rule N
o.3 is extracted. Antecedent of these two extracted rules
Of the data items in the leftmost column to which a value is assigned.
Data item (the second data item of Rule No. 1) and the rightmost data item
Data item (the eighth data item of rule No. 3)
The data item is divided into two at the position. For example, the first data
The fifth data item from the item is the left group, and the sixth data item is the
Seven data items are bisected into the right group. And that extracted
In each data item in the left group between the two rules
The values of the data items are exchanged with each other and two new rules
Is generated. That is, in the example of FIG.
The two new rules are as shown in FIG. If,
For all data items in the antecedent part of one of the generated rules
If no specific value is given, the rule is
Be extinguished. Thus, the second data item of the antecedent part
And a new rule with a value assigned to the seventh data item is generated
You. The conformity of the new rule is determined in step 316.
Is initialized to 0. Operations for generating such new rules
The operation changes the division position of the data item or is extracted
It is executed by arbitrarily changing the two rules. in this way
Then, many new rules are generated. like this
The rules obtained by simple operations are the characteristics of the two rules.
Has been taken over. Next, at step 318, the above
The rule base for which generational changes have been implemented is
Is stored in the fixed disk device 35. And the world
The rule base for which the substitution has been performed is shown in FIGS.
As described above, calculation of processing result data, correction input condition data
Prediction of the effect of machining results by calculation and correction input condition data
Used for calculations and the like. And in the process, each rule
The fitness is updated. Then update the database
When the new (generation change) time comes, the processing shown in FIG.
And the database is updated to the next generation of the database.
Be renewed. Such things are performed repeatedly
The database is ultimately an optimal rule base.
You. The optimal rule base is ideally the consequent part of the rule
For each rank value of each data item in
The rank value of the data item of the input condition data that affects
It is meant to be a single rule that is all optimized
To taste. According to the data exchange of the above two rules,
Is very similar to the two rules, and
New rules that combine the properties of
Is done. Good enough for such rule generation
High degree and the two previous benign components (causal relationship with conclusions)
Rules that are likely to take over comrades)
May occur. Inherited this benign club
A benign rule has a good relevance in subsequent real-world uses.
It is highly valued and will be
Survive even if you are. On the other hand, the inferiority of the two rules (conclusion and
Rule that inherits the weak condition factor)
Generated. However, its rule of inferiority is
Low conformity and eventually disappear from the rule base
Is destined to perish. In this way, the device of this embodiment is
In other words, rules can be self-propagating, self-differentiating, and self-evolving.
Thus, the rule base can be optimized. Yo
Therefore, the operation of adding a new rule by the operator is indispensable.
It is not necessary. In the above embodiment, a new rule is created.
If so, the value of any data item in any rule
May be generated by arbitrarily changing rules.
No. That is, generation of a new rule by crossover is 2
Since some of the characteristics of each of the rules are continued,
Rules do not change significantly
Base drops to energy minimum during optimization
Could be included. At this time, the value of the data item can be arbitrarily
By changing to random, so-called rule-based optimization
Cause disturbances in the process and sudden displacements in the rules
By causing it to escape from the energy minimum point,
To the minimum point of energy, that is, to complete the optimization
it can. In the above embodiment, the rule-based memory
The stage X1 is a rule of the fixed disk device 35 and the RAM 33.
The input condition data input means is constituted by a base area 335.
X2 is input condition data of the keyboard 22 and the RAM 33
Area 332, executed by CPU 31 and CPU 31
This is accomplished in steps 102-104 of FIG.
The processing result data predicting means X3 is configured to process the ROM 32
Result prediction program area 322, CPU 31 and CPU
Steps 108-112 of FIG. 8 performed by 31
Achieved by Further, the fitness calculating means X4 calculates the CP
U31 and the process of FIG.
Step 114─118, Step 212─216 in FIG.
Is achieved in. Further, the matching degree storage means X5 is provided in the RAM 33.
And a fixed disk device 35.
It is. The sorting means X4 is controlled by the CPU 31 and the CPU 31.
Performed in steps 302─310 of FIG.
You. Further, the rule generation means X5 includes the CPU 31 and the CPU 3
Step 312─316 of FIG. 10 performed by 1
Is achieved in.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a mechanism diagram showing a configuration of a numerically controlled grinding machine using a processing result prediction device according to a specific embodiment of the present invention. FIG. 2 is a block diagram showing a configuration of a numerical control device of the numerically controlled grinding machine of the embodiment. FIG. 3 is an explanatory diagram showing data items of rules. FIG. 4 is an explanatory diagram showing a data structure of a rule. FIG. 5 is an explanatory diagram showing a state of ranking of values of each data item. FIG. 6 is an explanatory diagram showing a rule configuration and a notation method. FIG. 7 is an explanatory diagram showing a procedure for creating a rule. FIG. 8 is a flowchart showing a processing procedure by a CPU for predicting processing result data using a rule base. FIG. 9 is a flowchart showing a processing procedure by a CPU for calculating input condition data from post-change processing result data using a rule base. FIG. 10 is a flowchart showing an update procedure by a rule-based CPU. FIG. 11 is an explanatory diagram illustrating how a new rule is generated by crossover in updating a rule base. FIG. 12 is an explanatory diagram showing a new rule generated by crossover. FIG. 13 is a block diagram showing a configuration of the present invention. [Description of Signs] 31 ... CPU 33 ... RAM 32 ... ROM

──────────────────────────────────────────────────続 き Continuing on the front page (72) Masaaki Minagawa, Inventor 2-20, Atsubetsu Chuo, 5-chome, 6-chome, Atsetsu-ku, Sapporo, Hokkaido (72) Inventor Takao Yoneda 1-1-1, Asahimachi, Kariya-shi, Aichi Prefecture (72) Inventor Masaru Yamanaka 1-1-1 Asahi-cho, Kariya-shi, Aichi Prefecture Inside Toyota Koki Co., Ltd. (56) References Patent No. 3337342 (JP, B2) (58) Fields investigated (Int. Cl. ⁷ , DB Name) B23Q 15/00 G05B 19/4155

Claims (1)

(57) [Claims] [Claim 1] An antecedent part in which input conditions consisting of machining conditions such as a feed rate and fixed conditions such as a workpiece finish dimension are described as preconditions when the rule is satisfied, A rule-based storage means for storing a plurality of rules including a consequent part in which a processing result is described as a conclusion when the rule is established, and a value in a range in which a plurality of rules can be processed as a plurality of values; Input condition data input means for inputting input condition data when the input condition data is input, and the antecedent portion of each rule stored in the rule base storage means for the input condition data input by the input condition data input means. A processing result data calculating means for obtaining processing result data based on the consequent part of the adapted rule; and the processing result calculated by the processing result data calculating means. A fitness value calculating means for inputting an evaluation value of data and calculating a fitness value indicating the fitness of the rule in accordance with the accumulation of the evaluation value; and a fitness value of each rule calculated by the fitness calculating device. A goodness-of-fit storage means for storing in correspondence with each rule; and a low-goodness rule stored in the rule-based storage means according to each goodness of each rule stored in the goodness-of-fit storage means. A selection unit that deletes and retains a rule with a high degree of conformity; and any two of the rules stored in the rule base storage unit, wherein the degree of conformity is greater than a predetermined value and the consequent parts are equal. A rule is selected, and a corresponding arbitrary number of the condition elements in the plurality of condition elements constituting the antecedent part are exchanged between the two rules to obtain 2
And a rule generation means for generating two new rules.