JP2007265233A - Process organizing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a process organizing method which reduces a period for organizing work by reducing a workload of a process organizing work, provides a uniform standard for the organizing work, and thereby prevents an omission in the process and reduces a processing period. <P>SOLUTION: The process organizing method, wherein a production process comprises a plurality of processes and a plurality of works are organized so that the plurality of works allocated to each of the plurality of processes are not exceed a predetermined number of man-hours, comprises steps of: performing the approximate optimization search 31 comprising a genetic algorithm 21 and a hill-climbing algorithm 22; and pre-search processing 33 prior to the approximate optimization search. In the pre-search processing 33, a combination of work shift patterns is optimized (step S47), and a combination of whole patterns is determined (step S48). <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a process knitting method, and more particularly to a process knitting method suitable for supporting knitting of a production process in a production process in which a plurality of types of products are appropriately combined on a production line.

  For example, in the field of production of various industrial products, in recent years, from the viewpoint of cost competition, there is a strong demand for optimization of product production efficiency and production line operation rate. In addition, as a result of diversifying user demands and preferences, various small-volume productions are performed to meet such demands, and production line processes are frequently changed at production sites. For this reason, the operation rate is high, and quick response of an optimal production process is required.

  In the production process organization, production tact fluctuations occur every month due to fluctuations in total production units and production models. When new models are introduced, it is necessary to create a new production process organization.

  There are many factors in the organization of production processes, such as the flow time of the models to be produced, the assembly order of parts, the number of work steps, the presence / absence of dedicated equipment / carts, work position, difficulty of work, and attendance / experience of workers. It is related and complicated. For this reason, it takes a lot of labor and time to reorganize the production process, and this is usually limited to a few skilled personnel. In addition, since each person in charge organizes the production process individually by hand, the creation standards are not uniform, and there is a possibility that the omission of the process may occur.

  If the creation of the production process organization is delayed, advance preparation and process training at the production site become insufficient, which may cause a vicious circle. For this reason, it is required to organize and standardize the production process organization standards.

  However, since the combinations of production processes increase exponentially according to the complexity of the production process organization, it is difficult to find all combinations brute force to find the optimal combination due to constraints such as work time. It is.

There is an optical axis adjustment method described in Patent Document 1 as one of conventional techniques for obtaining an optimal solution when the number of combinations is enormous. As described in Patent Document 1, optical axis adjustment with multiple degrees of freedom for optical components has a huge number of combinations as in the production process organization, but the genetic algorithm and the hill-climbing algorithm are combined. Thus, the optimization search is performed approximately. Thereby, the optical axis adjustment time can be shortened.
Japanese Patent No. 3645168

  The genetic algorithm is a global search method when there are many local peaks, that is, local solutions, and the processing speed is faster than the hill-climbing method when there are many local solutions. However, at the end of the search for the solution by the genetic algorithm, the search speed decreases, and the processing speed to reach the local solution is slower than the hill-climbing method. Therefore, in Patent Document 1, the processing time is shortened by a technique of first adjusting the optical axis using a genetic algorithm and then switching to the hill-climbing method as a fine-tuning search. Thereby, workability can be remarkably improved without requiring highly precise manual adjustment by a skilled person.

  However, even the search method for the combination of the genetic algorithm and the hill-climbing method employed in the optical axis adjustment method described in Patent Document 1 has a limit to process an enormous number of combinations, and the true local There was a problem that the search ended without reaching the solution.

  More generally, in the search method, a simple one-dimensional individual (chromosome) is used for the search process. However, in this production process organization, the individual (chromosome) becomes large, and thus a lethal gene is likely to occur. The problem of increased processing time is raised.

  In view of the above problems, the object of the present invention is to perform the process knitting work without depending on the skill level of the person in charge, reduce the labor burden of the process knitting work, shorten the knitting work time, A uniform standard can be given to the knitting work, which can prevent the occurrence of omissions in the process, reach the true local peak as much as possible, and further suppress the occurrence of lethal genes to select individuals (chromosomes) It is an object of the present invention to provide a process organization method that can be made compact without waste and that can shorten the processing time.

  In order to achieve the above object, the process organization method according to the present invention is configured as follows.

In the first process knitting method (corresponding to claim 1), the production process is formed from a plurality of processes, and the plurality of works are arranged so that the plurality of works arranged in each of the plurality of processes is equal to or less than a predetermined man-hour. The process organization method includes the following data setting stage, pre-search process stage, initial individual population generation stage, and search process stage.
In the data setting stage, process data in which position information of each of a plurality of operations arranged in each of a plurality of processes is recorded, and whether each of the plurality of operations is adapted to the work conditions of the production process is evaluated. Fitness data recording work condition information is set.
In the pre-search processing stage, a movement pattern for transposing an arbitrary work to another process is generated as a work movement pattern so that each of the plurality of works on the process data has a predetermined man-hour or less in the arranged process. Determining whether or not the number of combinations of work movement patterns, which are all aggregates of the work movement patterns, exceeds a first threshold; and the number of combinations of work movement patterns sets the first threshold If it exceeds, based on the evaluation value of the work movement pattern calculated from the fitness data, a predetermined number of work movement patterns with low evaluation values are deleted, and then a combination of all patterns that are all aggregates of the work movement patterns Generating.
The initial individual population generation stage is a part of the processing contents included in the genetic algorithm, and a step of selecting a work movement pattern under a predetermined condition in the combination of the whole pattern and extracting and connecting only the selected work movement pattern. A step of generating an individual by a genetic algorithm, and a step of repeating the step of generating an individual by a genetic algorithm a predetermined number of times. These steps generate an initial population of genetic algorithms.
The search processing stage includes a first search step for searching for a plurality of work arrangements according to a genetic algorithm based on the initial individual population, and a second search for further searching for a plurality of work arrangements according to a hill-climbing method after the first search step. And a step of repeating the first search step and the second search step.
The approximate optimization search process is constituted by the initial individual population generation stage and the search processing stage.
In the first process composition method described above, the fitness of process composition is set in advance in the data setting stage, and the combination of the work movement patterns is deleted from the setting in the pre-search processing stage to perform optimization. Then, a combination of overall patterns of process organization is calculated and determined. An initial population is generated from the combination of the whole patterns calculated in the pre-search process, and then an approximate optimization search is performed by combining the genetic algorithm and the hill-climbing method. By providing the pre-search processing stage, the number of combinations of work movement patterns is made appropriate, and the processing time can be shortened.

  The second process organizing method (corresponding to claim 2) is preferably the above-described process organizing method, preferably in the pre-search processing stage, when the number of combinations of the entire patterns does not exceed the second threshold value. It is characterized by executing a sequential search process step for searching for a pattern. In this process organization method, switching to approximate optimization search or all pattern search is performed according to the number of combinations of all process organization patterns calculated in the search pre-processing, thereby searching for the optimum process organization.

  According to a third process organization method (corresponding to claim 3), in the above-described process organization method, preferably, an initial individual population for executing a genetic algorithm is generated, and a work movement pattern for each of a plurality of works is provided. Is formed as a gene sequence, a gene sequence for each of a plurality of steps is formed as a gene sequence, and the gene sequence is formed into a jug sequence arranged as rows in the order of the steps. In addition, this jagged sequence is generated by forming one gene from a movement destination candidate and work contents, and arranging the gene sequence in the order of processes.

The process organization method according to the present invention has the following effects.
According to the process organization method according to claim 1, in the process organization method that combines a genetic algorithm and a hill climbing method that is executed thereafter, and repeats this to execute the search processing stage, Since the fitness data is set, and based on these data, the search pre-processing stage for optimizing the number of combinations of the overall patterns related to the combination of work movement patterns is provided before the search processing stage is executed. The organization of the process is not limited to skilled personnel, but further suppresses the generation of lethal genes and makes the individual (chromosomes) compact without waste, so it can be searched and processed in a short time. In addition, the knitting standards for the knitting work of the production process are unified, it is possible to prevent the omission of the process, and even if the total number of combinations of the whole pattern is enormous, It can be searched treated in degrees and high speed.
According to the process organization method according to claim 2, since the approximate optimization processing and the entire pattern search are switched even if the number of the combinations of the entire patterns is small, the optimum according to the number of combinations of the entire patterns. Search can be performed.
According to the process organization method according to claim 3, since the generation of lethal genes can be suppressed in advance and the genes of the initial individual population can be made compact without waste, the number of lethal genes is reduced, with high accuracy and high speed. Search processing can be performed.

  DESCRIPTION OF EMBODIMENTS Preferred embodiments (examples) of the present invention will be described below with reference to the accompanying drawings.

  A representative embodiment of the process organization method according to the present invention will be described with reference to FIGS. 1 to 12 and Tables 1 and 2 below. In this embodiment, description will be made assuming a production process of, for example, an automobile or a motorcycle. However, the production process to which the present invention is applied is not limited to such products.

  As an example, in a production process consisting of one production line, the production process is composed of a plurality of (for example, “p”) processes 1 to p, and at least one required for each of the processes 1 to p. One operation is carried out. The amount of work performed in each of the steps 1 to p is determined under the restriction that it does not exceed the range of appropriate man-hours. A number of processes are set in the production process of an automobile, etc., and at least one operation required in each process is performed. In the process organization method of the production process, a number of operations performed in each of the plurality of processes 1 to p are appropriately determined. In particular, when producing many types of products on one production line, it is required to quickly perform an optimal process organization when changing monthly conditions or changing production models.

  The process organization method according to the present embodiment uses a genetic algorithm and is configured to include this. First, the genetic algorithm is outlined.

  The genetic algorithm has “selection selection”, “crossover”, and “mutation” as basic operations, and it is necessary to set genotypes, initial gene populations, and fitness evaluation criteria in advance. This genetic algorithm is an effective and widely applicable technique for searching for combinations having such a large solution space that it is considered impossible to perform a full search. On the other hand, however, the genetic algorithm may not give a good result depending on what the target is parameterized as a gene and what evaluation function is used to evaluate the fitness.

  In a general genetic algorithm, a group of hypothetical individuals called chromosomes is set first, and individuals that adapt to a predetermined environment survive according to their high fitness and leave offspring. Try to increase the probability. Then, the parent gene is inherited by the child through a procedure called genetic manipulation, that is, the above-mentioned “selection”, “crossover”, and “mutation”. By carrying out such generational changes and evolving genes and individual populations, individuals with high fitness will occupy the majority of the individual populations. Genetic crossovers, mutations, and the like, which are genetic operations at that time, mimic gene manipulations that occur in the reproduction of actual individuals.

  FIG. 1 shows a flowchart of a search based on a general genetic algorithm.

  In the first step S11, an initial population of individuals, that is, n chromosomes is determined, and what kind of data is transmitted in what format from the parent individual to the offspring individual during the generation change. . In the genetic algorithm, at the start of the search, the target problem is generally a complete black box, and it is unclear what kind of individual is desirable. For this reason, the initial individual population in step S11 is normally generated randomly using random numbers. Therefore, since a bad gene can also be a candidate, if there is some prior knowledge about the search space, processing such as generating an individual population centering on a portion that is considered to have a high evaluation value may be performed. The total number of individuals to be generated is referred to as “the number of individuals in the group”.

  An example of a general one-dimensional individual group is shown in FIG. In FIG. 2, n chromosomes are shown in the vertical direction. Each of these n chromosomes is a fixed-length chromosome composed of m genes ((1), (2),..., (M)) as shown in the horizontal direction. Each of the m genes is determined based on a certain integer, real number, symbol or the like. A set of these m genes means one individual, that is, one chromosome.

  In this embodiment, as will be described later, since the gene is set as a variable length, the individual (chromosome) has a jug sequence.

  In the stage before the transition from step S11 to the next step S12 (calculation of fitness), “fitness” indicating how much each of the n individuals is adapted to the set environment. "Evaluation table" is determined. In subsequent generational changes, individuals with higher fitness for the environment are made more likely to survive or create offspring than other individuals with lower fitness. Conversely, individuals with a low degree of fitness for the environment are considered to be individuals that are not well adapted to the environment and disappear.

  In this embodiment, the “fitness evaluation table” shown in Table 1 below is prepared. Based on the “fitness evaluation table” shown in Table 1, each evaluation criterion is scored, and the fitness of each individual is calculated. In this evaluation table, items of “Must / Want”, “Evaluation criteria”, “Appropriate value”, and “Evaluation unit” are set on the horizontal axis.

  In the “fitness evaluation table” shown in Table 1, “Must” means a condition that must be cleared, and if it cannot be cleared, it becomes a lethal gene. “Want” means a desirable condition. According to the “fitness evaluation table” in Table 1, the knitting standards are unified, and the knitting work of the production process is not limited to skilled personnel, and it is possible to prevent omission of the process.

  In the next step S12 in the flowchart shown in FIG. 1, the fitness of each individual in the individual population is calculated based on the “fitness evaluation table”.

  After the fitness for each of the n individuals is calculated in step S12, whether or not the next-generation individual population generated in step S13 satisfies the evaluation criteria for completing the search. Check out.

The above evaluation criteria are determined depending on the problem, but typical ones are as follows.
(A) The maximum fitness in the individual population is greater than a certain threshold.
(B) The average fitness of the entire population is greater than a certain threshold.
(C) The increase rate of the fitness of the individual group continued for a certain period or more for the generation below a certain threshold.
(D) The number of generation changes (number of generations a) has reached a predetermined number (set value).

  In the genetic algorithm in this embodiment, (D) is adopted as an evaluation criterion. Specifically, for example, the search is terminated when the number of generation changes (generation number a) is 300 generations.

  If the search termination condition is satisfied in the above step S13 (in the case of YES), the process proceeds to step S18 to end the search, and the optimization problem for finding the individual with the highest fitness in the individual population at that time The solution of If the search termination condition is not satisfied in step S13 (NO), the process proceeds to step S14, and a series of steps S14 to S16 are performed to perform the search. After step S16, the generation number (a) is incremented by 1 in step S17, and the process returns to step S12. In the case of NO in step S13, the search is continued by repeating such generation change, and the fitness of the individuals is increased while keeping the number of individuals in the group constant.

  When the search termination condition is not satisfied, in step S14, an individual to be a basis for the next generation of individuals is selected from the group. However, simply performing selection will only increase the proportion of individuals having the highest fitness at the present time in the population, and no new search points will be generated. For this reason, operations called “crossover” in step S15 and “mutation” in step S16 are performed.

  In the “selection kite” in this embodiment, based on the “fitness evaluation table” shown in Table 1 above, the “elite strategy method” is implemented to ensure that the predetermined number of individuals with the highest evaluation is left to the next generation. Then, in order to keep the number of individuals in the group constant, the “tournament method” is used to select the remaining individuals by comparing two randomly selected individuals and adopting individuals with high evaluation values as needed. carry out.

  Individuals that have been adopted and those that have not been adopted in a single comparison are returned to candidates and again adopted as targets for the tournament method. The total number of individuals selected by the elite strategy method and tournament method is the number of individuals in the group. In the genetic algorithm of this embodiment, the number of individuals in the group is set to 200, for example.

  In step S15, a pair of two individuals is randomly selected at a predetermined frequency from the next generation individuals generated by the “selection kite”, and chromosomes are rearranged to create child chromosomes (crossover). Here, the offspring individuals generated by “crossover” are individuals that have inherited the traits from each of the individuals that are parents. This crossover process increases the diversity of individuals' chromosomes and causes evolution.

  In the “crossover” of the present embodiment, a general uniform crossover is performed. An example of uniform crossover is shown in FIG. In FIG. 3, mask pattern 11, chromosomes A and B before crossover, and chromosomes A and B after crossover are shown. Chromosomes A and B before crossing are individuals who are parents, chromosome A has the same gene as chromosome 1 shown in FIG. 2, and chromosome B has the same gene as chromosome 2 shown in FIG. Chromosomes A and B after crossing are individuals corresponding to offspring. According to the uniform crossover, as shown in FIG. 3, for the gene of the child chromosome A, the gene of the parent chromosome A is copied at the position where the mask pattern is “0” by the mask pattern 11 prepared in advance. At the position where the mask pattern is “1”, the gene of parent B is copied. Similarly, with respect to the gene of child chromosome B, the gene of parent chromosome B is copied at the position where the mask pattern is “0” and the parent A is generated at the position where the mask pattern is “1”. Copy the gene. This uniform crossover is considered a kind of multipoint crossover. In the present embodiment, the mask pattern 11 is randomly generated every time crossover is performed.

  After the crossover process, the gene of the individual is changed with a predetermined probability by the mutation in the next step S16. Here, the phenomenon that the content of a gene is rewritten with a low probability is a phenomenon that is also observed in the gene of an actual individual. However, it should be noted that if the gene content is changed too much, the inherited characteristics of the parental traits due to crossover are lost, which is similar to a random search in the search space. In the mutation according to this embodiment, mutation is performed at a mutation rate of 1% in each chromosome, and the value of the gene is randomly changed with a random number.

  After execution of step S16, the process returns to step S12 via step S17. In step S17, the number of generations increases by one and the generation advances.

The next generation population is determined by the above processing. This is a general genetic algorithm outline and does not prescribe a detailed algorithm. In order to apply the genetic algorithm to this embodiment, it is necessary to appropriately set the following items (A) to (F).
(A) Initial population setting (B) Fitness setting (C) Selection method (D) Crossover method (E) Mutation method (F) Search termination condition

  In the case of the genetic algorithm applied to the present embodiment, in practice, the uniform crossover is partially changed as will be described later.

  In the present embodiment, the various parameters (A) to (F) (setting values, threshold values, etc.) are tuned in advance by simulation, and this is obtained using a general meta-GA technique. “Meta-GA method” is a method of creating a macro model and optimizing various parameters used in the genetic algorithm, and extracting the optimal parameters using the genetic algorithm using the above parameters as genes. .

  Further, in this embodiment, an approximate optimization search algorithm is used in which the above-described genetic algorithm is further combined with a hill-climbing search method. A flowchart of this approximate optimization search algorithm is shown in FIG. In FIG. 4, block 21 indicates the genetic algorithm described in FIG. 1, and the same reference numerals are given to the same elements as the steps described in FIG. 1 in block 21. After the mutation step S16, the hill-climbing algorithm block 22 is added. The hill-climbing algorithm block 22 includes step S21 of performing the hill-climbing method. After step S21 is executed, the process returns to step S12 of the genetic algorithm 21 via step S17.

  The genetic algorithm is a global search method when there are many local peaks, and the processing speed is faster than the hill-climbing method when there are many local peaks. However, the search speed drops at the end of the search, and the processing speed to reach the local peak is slower than the hill-climbing method. The hill-climbing method is a general method for guiding and searching only in the direction with good evaluation. For this reason, the search may end at a certain local peak, and the true peak may not be reached. Therefore, the initial group is selected and crossed by a genetic algorithm, and a hill-climbing search is performed in parallel. In addition, by causing random changes from time to time due to mutation, it is possible to escape from local stability (local optimization).

  In this embodiment, the hill-climbing method in step S21 is performed by mutating individuals with poor evaluation according to the “fitness evaluation table” shown in Table 1 in the direction of high evaluation with high probability, and using them together with the genetic algorithm. Improve search efficiency. In the hill-climbing method of the present embodiment, mutation is performed with high probability in each chromosome (1 to n), and the value of the gene is randomly changed with a random number. Try to leave.

  As will be described later, when there are few combinations of the number of production processes, an entire pattern search (sequential search) for searching all patterns is performed without performing an approximate optimization search. When all pattern search is used in this way, 100% search is possible. This is illustrated in FIG.

  Next, based on the flowchart shown in FIG. 5, the search algorithm of the process organization method which concerns on this embodiment is demonstrated.

  In FIG. 5, a block 31 means a portion for executing an approximate optimization search including the genetic algorithm 21 and the hill-climbing algorithm 22 described in FIG. The block 32 means a part for executing the full pattern search when the combination of the number of production processes is small. The block 32 for executing the full pattern search includes the step S31 of the sequential search.

  In the process knitting method according to the present embodiment, the combination of production process knitting exponentially increases in accordance with the complexity of the production process knitting. For this reason, it is difficult to find an optimal combination by searching all combinations in a brute force manner. In addition, since the search time is limited, if the number of combinations increases, the search may end halfway and the search accuracy may deteriorate.

  Therefore, as shown in FIG. 5, before the approximate optimization search 31 or the full pattern search 32 is performed, the search pre-processing 33 for optimizing the number of combinations of the number of formations in the production process is performed in advance. Combinations that cannot be expected are deleted to improve the high speed and accuracy of the approximate optimization search 31 and the like. This means that when the same 300 generations are processed by the approximate optimization search 31, combinations that cannot be expected in advance are not included, thereby suppressing the occurrence of lethal genes and eliminating unnecessary processing. I mean.

  In the flowchart shown in FIG. 5, in the first step S41, before performing the process organization process, the work of the process with the appropriate man-hour or more is confirmed and the search is started. In the next step S42, a method for calculating the fitness of the process organization is set in advance. Thereafter, the process proceeds to the search preprocessing 33.

  In the process of search pre-processing 33, in the first step S43, the work destination candidate is confirmed from the viewpoint of “where each work can move”.

  An example of the above steps S41 to S43 will be specifically described with reference to FIG. First, in step S41, the process of the process more than an appropriate man-hour is confirmed, and a search is started. FIG. 6 shows a state (A) before the process organization and a state (B) after the process organization. In the state (A) before the process organization, the horizontal axis means the process axis along the time axis, and for example, the processes 1 to 6 are shown. The vertical axis represents the man-hour (type and number of work) assigned in each of the steps 1 to 6. In the state (A) before the process organization, works A, B, C, and D are set in process 1, works E, F, and G are set in process 2, and works H, I, J, K, and so on are set in process 3. L and M are set. In the state (A) before the process organization, the line 34 means an appropriate man-hour. Considering the line 34 with the appropriate man-hours as a reference, the processes 1 to 3 exceed the line 34 with the appropriate man-hours. Therefore, it is necessary to move the work to another process so that the man-hours become appropriate.

  Next, in step S42, a method for calculating the fitness of the process organization is set in advance, and then in step S43, work destination candidates are confirmed. For example, in steps 4, 5, and 6 as work destination candidates, for example, operations D, G, I, J, K, and more, which are greater than the appropriate man-hours, as indicated by the arrow 35 in the state (A) before the process organization. L and M are moved to the subsequent steps 4 to 6. Also in this case, the conditions of the line 34 with the appropriate man-hour are satisfied in each of the steps 4 to 6. Therefore, in the example of the state (B) after the process organization in FIG. 6, as an example, the work G is moved to the process 4, the works D, I, and M are moved to the process 5, and the works J and K are moved to the process 6. , L. As described above, it is necessary to move the work in the processes 1 to 3 that is equal to or more than the appropriate man-hour to each of the processes 4 to 6 as in the state (B) after the process organization. In order to perform such movement, in step S43 described above, from the viewpoint of “where each work can move,” “impossibility of work movement destination” is set as the fitness evaluation table (work (Restriction of characteristics) In Table 2 below, the relationship between the above-described operations A to M and the steps 4 to 6 which are candidate work movement destinations, that is, the above-described work movement pattern is generated, and is indicated by the second symbol “◯”. The performed process is a process in which each of the operations A to M can be moved. For this reason, by not including combinations that cannot be expected in advance, the generation of lethal genes is suppressed, and the search processing time can be shortened with high accuracy.

  In the example shown in FIG. 6, the number of work movement destination candidates is relatively small, but in reality, many work movement destinations are candidates.

  In the next step S44, a work movement pattern is calculated for a number of work movement destination candidates obtained in step S43 for confirming the work movement destination candidates. Here, on the assumption of the example shown in FIG. 6, in order to make the number of steps 1 to 3 exceeding the appropriate number of steps appropriate, it is assumed that the work movement pattern is assumed to move to steps 4 to 6 which are work transfer destinations. To all tables. At this time, as shown in FIG. 7, on the basis of the fitness evaluation table shown in Table 1, the combinations with good evaluation are arranged in order from the top.

In the next step S45, calculation of a combination of work movement patterns is executed. Based on the work movement pattern calculated in step S44 described above, the number of combinations of work movement patterns is obtained by multiplying the number of work movement patterns (PN1) in steps 1 to 3 in FIG.
Combination of work movement patterns = step 1 (PN1) × step 2 (PN1) × step 3 (PN1)
= 3 × 1 × 6
= 18 (street)

  FIG. 7 shows a table (A) indicating work movement patterns in each process and a table (B) indicating the number of work movement destination candidate patterns. According to the table (A) in FIG. 7, as an example, the area 41 shows a combination of work movement patterns related to the process 1, the area 42 shows the combination of work movement patterns related to the process 2, and the area 43 shows the work movement related to the process 3. Indicates a combination of patterns. The number of work movement patterns in process 1 is 3, the number of work movement patterns in process 2 is 1, and the number of work movement patterns in process 3 is 6. Thereby, the number of combinations of work movement patterns given by the above equation is obtained.

  In the next determination step S46, it is determined whether or not the number of combinations of work movement patterns obtained in step S45 is larger than a preset first threshold value. If the number of combinations of work movement patterns is larger than a preset first threshold value (in the case of YES), a combination with poor evaluation is deleted from the fitness evaluation table of Table 1 in step S47. Step S47 is a step of optimizing the combination of work movement patterns. In this embodiment, the first threshold value is set to “5”, for example, and one work from step 3 (the number of work movement patterns is 6) having more combinations of work movement patterns than the first threshold value “5”. Delete the movement pattern. Actually, in FIG. 7, the work movement pattern 6 is deleted from the work movement patterns 1 to 6 in step 3. After execution of step S47, the process proceeds to step S48.

  If the number of combinations of work movement patterns is smaller than the preset first threshold value in the determination step S46 (NO), the process proceeds to step S48. In step S48, the overall pattern combination is determined.

  For example, the work B and work C move destination candidates that are the work move contents of the work move pattern combination 3 (shown in the area 41 of FIG. 7) obtained in step S44 are shown in Table 2 used in step S43. As described above, the operation B is obtained as two ways of steps 4 and 6, and the operation C is obtained as three ways of steps 4, 5, and 6. Therefore, the number of work movement destination candidate patterns (2 × 3 = 6 patterns) is added as the number of work movement destination candidate patterns to the table part ((B) of FIG. 7) of the work movement pattern created in step S44.

  Similarly, for other work movement patterns, the number of work movement destination candidate patterns is calculated for each pattern, and the number of movement work destination candidate patterns is calculated for all of the entire work movement patterns. In this case, as described above, the combination of work movement patterns is optimized in step S47, and combinations that cannot be expected to be effective are deleted in advance. The reason is as follows.

Hereinafter, with respect to the example shown in FIG. 7, the number of combinations of the entire patterns when the pattern 6 in the work movement pattern of the process 3 is deleted and when not deleted is shown. The total number of combinations of patterns can be obtained by multiplying the number of candidate work movement destination patterns (PN2) in steps 1 to 3.
(1) When deleting pattern 6:
Combination of whole patterns = step 1 (PN2) × step 2 (PN2) × step 3 (PN2)
= (3 × 2 × 6) × (2) × (2 × 24 × 36 × 36 × 24)
= 107,495,424 (street)
(2) When pattern 6 is not deleted:
Combination of whole patterns = step 1 (PN2) × step 2 (PN2) × step 3 (PN2)
= (3 x 2 x 6) x (2) x (2 x 24 x 36 x 36 x 24 x 36)
= 3,869,835,264 (street)

  As described above, the number of combinations of the whole patterns greatly varies depending on whether or not the work movement pattern 6 in step 3 is deleted. If the pattern 6 in the step 3 is not deleted, the number of combinations of the whole patterns becomes enormous and the amount of calculation becomes large. Therefore, the combinations of work movement patterns are optimized.

  As described above, when the number of combinations is smaller than the first threshold value in the determination step S46, the process directly proceeds to step S46, and when the number of combinations is larger than the first threshold value, the process proceeds through the step S47 for optimization of the combination. Then, the process proceeds to step S48. In step S48, a combination of overall patterns is determined.

  Thereafter, the process proceeds to step S49. In step S49, when the calculated combination of the whole patterns is larger than the preset second threshold value, an approximate optimization search 31 including the steps of the genetic algorithm 21 and the hill-climbing method 22 is performed. If the calculated combination of the entire patterns is smaller than the second threshold value in step S49, an all pattern search 32 including step S31 is performed.

  In the approximate optimization search 31, in the step of the genetic algorithm 21, an initial individual (chromosome) group is first randomly generated from the entire pattern calculated by the search preprocessing 33 using random numbers. FIG. 8 shows specific processing contents relating to generation of an individual group. The processing process shown in FIG. 8 includes two steps S51 and S52. In step S51, a predetermined gene is randomly selected under the relationship table between the work movement patterns of each of steps 1, 2, and 3 and the movement destination candidates. Step S52 is a return step for returning to the beginning of step S51. In step S11 of the genetic algorithm 21 shown in FIG. 4, steps S51 and S52 are executed.

  As means for generating an initial individual group, in step S51, one pattern for each process of the entire pattern is selected at random, and a destination candidate is randomly determined for each work content in the pattern. If a work movement pattern and a movement destination candidate are selected for each process, the selected ones are combined into one. In addition, by the return processing in step S52, step S51 is repeated a predetermined number of times to generate an initial individual population. In this embodiment, for example, by repeating 200 times, an initial individual population having 200 chromosomes is generated.

  The individual (chromosome) of the production process organization generated according to FIG. 8 becomes a jagged array by patterning the work movement from step S42 to step S44. In addition, since this individual (chromosome) is compact and does not easily include combinations that cannot be expected in advance, the search processing time can be shortened with high accuracy.

  Here, an image of the initial population (chromosomes) is shown in FIG. In this embodiment, each individual (chromosome) of the individual population has a gene consisting of a jug sequence. In FIG. 9, steps 1, 2, 3 and genes (1), (2), (3), (4) are shown for chromosomes 1, 2,..., N. Genes consist of “work contents and destination candidates”. The work contents and destination candidates are variable-length dynamic arrays, and the number of processes is also variable-length dynamic arrays.

  In the genetic algorithm applied in the present embodiment, the individual (chromosome) is a jagged array, but as described later, crossover (uniform crossover) can be performed when the processes match.

  Here, as shown in FIG. 10, when the comparison is made between the chromosome 1 and the chromosome n, the point where the steps are coincident is the step 3, so the crossover of the step 3 is performed. There are three genes that can be crossed in “Step 3”: “I →”, “J →”, and “K →” included in each of the block 51 and the block 52 having the same work movement content. Each is crossed by a mask pattern 11 set at random, and the genes are exchanged.

  When the number of individuals (number of chromosomes) is too large, the load becomes high and the processing becomes slow. If the number of individuals (number of chromosomes) is too small, the number of multipoints to be searched is reduced, and the processing is slowed down. For this reason, the optimal number of individuals (number of chromosomes) is obtained in advance by the meta-GA technique.

  In the process organization method according to the present embodiment, a process organization search test was performed based on the following search test conditions, thereby obtaining a search test result shown in FIG. In the graph of FIG. 11, the horizontal axis represents the number of generations, and the vertical axis represents fitness. A graph 61 shows a characteristic that the fitness changes as the number of generations increases. The line 62 is a line that does not violate the Must condition in Table 1 described above, the line 63 is a line that satisfies the Want condition, and the line 64 is a substantially optimal line. From this search result, it can be seen that a practical approximate solution and processing time can be obtained with this search method regardless of the size of the solution space.

<Search test conditions>
Search method Search pre-processing + approximate optimization search (genetic algorithm + hill-climbing method)
Process organization pattern Process increase pattern Parameter tuning Meta GA method Number of individuals (number of chromosomes) 200 Evaluation of fitness Calculated according to Table 1 Selection selection method Elite strategy method + tournament method Crossover method Uniform crossover Mutation method Mutation (normal) Mutation) + mountain climbing method (mountain climbing mutation)
Search termination condition When 300 generations are reached Solution space There are about 10 ²⁰ combinations of the entire pattern

  FIG. 12 shows the configuration of an apparatus (process knitting apparatus) for carrying out the process knitting method according to the present invention. This process organization apparatus 70 is constructed by a computer 71. The internal storage device built in the computer 71 stores a program 72 for executing the process organization method of the present invention. The program 72 realizes the function of the process organization calculation unit 73 that implements the process organization method. Further, in the program 72, a work movement destination candidate part 74 for storing the created work movement destination candidates, an overall pattern part 75 for storing the created whole pattern, and an individual group part 76 for storing the created individual group. Is included. An input unit 77 such as a keyboard or a mouse, a display unit (monitor) 78, a printer 79, and an external storage device 80 are attached to the computer 71. The external storage device 80 includes, for example, an fitness data recording unit 81 and a process data storage unit 82.

  The configurations and the like described in the above embodiments are merely shown to the extent that the present invention can be understood and implemented. Therefore, the present invention is not limited to the described embodiments, and Various modifications can be made without departing from the scope of the technical idea shown in FIG.

  The process organization method according to the present invention is used for process organization in a production process of an automobile, a motorcycle, or the like.

It is a flowchart for demonstrating a genetic algorithm. It is a figure for demonstrating a chromosome. It is a figure for demonstrating the crossing between two chromosomes. It is a flowchart which shows how to combine the genetic algorithm and hill-climbing method which are employ | adopted with the process organization method which concerns on embodiment of this invention. It is a flowchart which shows the whole structure of the process organization method which concerns on embodiment of this invention. It is a figure for demonstrating the example of the work movement pattern in the process organization method which concerns on this embodiment. It is a figure for demonstrating the content (A) of work movement in each process, and the number (B) of work movement destination candidate patterns. It is a figure for demonstrating the creation order of an individual | organism | solid. It is a figure explaining the example of the chromosome (individual population) used with the genetic algorithm in the process organization method which concerns on this embodiment. It is a figure explaining the example of the crossing by the genetic algorithm in the process organization method which concerns on this embodiment. It is a graph which shows the search test result to which the process organization method concerning this embodiment is applied. It is a block diagram of the process organization apparatus for enforcing the process organization method which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Mask pattern 21 Genetic algorithm 22 Mountain climbing method 31 Approximate optimal search 32 All pattern search 33 Search pre-processing 70 Process organization apparatus

Claims (3)

A production process is formed from a plurality of processes, and a plurality of processes arranged in each of the plurality of processes is a process knitting method for knitting the plurality of works so that the number of work is less than or equal to a predetermined man-hour.
Process data in which position information of each of the plurality of operations arranged in each of the plurality of processes is recorded, and an operation for evaluating whether each of the plurality of operations is adapted to a work condition of the production process. A data setting stage for setting fitness data recording condition information;
Generating a movement pattern as a work movement pattern for transposing an arbitrary work to another process so that each of the plurality of works on the process data is equal to or less than the predetermined man-hour in the arranged process; Determining whether the number of combinations of the work movement patterns, which are all aggregates of the work movement patterns, exceeds a first threshold, and the number of combinations of the work movement patterns is the first When the threshold value is exceeded, based on the evaluation value of the work movement pattern calculated from the fitness data, the predetermined number of the work movement patterns having a low evaluation value are deleted, and then all the groups of the work movement patterns are collected. A pre-search stage including a step of generating a combination of certain global patterns;
Selecting the work movement pattern under a predetermined condition in the combination of the whole patterns, extracting and connecting only the selected work movement pattern, generating an individual by a genetic algorithm, and an individual by the genetic algorithm Generating an initial population of the genetic algorithm, thereby generating an initial population of the genetic algorithm.
A first search step for searching for the organization of the plurality of tasks according to the genetic algorithm based on the initial individual population, and a second search step for further searching for the organization of the plurality of tasks according to a hill climbing method after the first search step. And a search processing stage including the step of repeating the first search step and the second search step,
The process organization method characterized by comprising.   2. The sequential search processing step of searching for all patterns is executed when the number of combinations of the whole patterns does not exceed a second threshold value in the pre-search processing stage. The process organization method.   In the generation of the initial population for executing the genetic algorithm, the work movement pattern for each of the plurality of work is set as one gene, and the gene for each of the plurality of steps is formed as a gene string. 3. The process organization method according to claim 1, wherein the gene sequences are formed in a jug arrangement in which the gene sequences are arranged as rows in the process order.

Images