JP4196764B2 - Cause / route estimation method and cause estimation apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention is configured in such a manner that a plurality of components to be processed are arranged in accordance with a predetermined flow in order to generate a flow from the upstream side to the downstream side of the processed element. In particular, the present invention relates to a technique for analyzing a cause that causes an idle state in a certain component, and in particular, a cause element that is another component that causes an idle state in a certain component and the cause element to be in the idle state The present invention relates to a technique for estimating a cause route to an idle component that is a component.
[0002]
[Prior art]
There is already a system in which a plurality of components to be processed are arranged according to a predetermined flow in order to generate a flow from the upstream side to the downstream side of the processed element. To do. An example of this type of system is a production system that produces products by processing parts or combining them with other parts.
[0003]
The simplest method for improving the processing capacity (throughput) of this production system is to allocate a buffer to the production system. This is because buffer allocation allows for rapid changes in the system and requires less initial cost for the change. The initial cost is the cost of adding or removing buffers, which is usually only part of the cost of adding a processing machine or changing the flow of the system.
[0004]
However, increasing the buffer size will always increase throughput, but sometimes the in-process inventory and total processing time will also increase, resulting in increased inventory and time to reach the consumer. Therefore, an important interest is gathered in research on the trade-off between the processing capacity, the number of in-process inventory and the total processing time.
[0005]
Buffers can improve the availability of other machines by reducing the idle time of other machines (the time the machine was starved or blocked) due to the machines that make up the bottleneck in the system. The state where the machine is starved (the parts are deficient) means that the machine cannot operate due to the lack of parts supplied to the machine, while the machine is blocked (closed). Means a state in which the machine cannot work because the machine cannot carry out the processed parts.
[0006]
The buffer serves two purposes in particular. That is, the buffer reduces the machine staging time by adding parts, and reduces the machine blocking time by adding free space, thereby improving the throughput of the system.
[0007]
There are many search methods for buffer allocation. Most of these methods, for example, use simulated annealing methods, genetic algorithms, neural networks, gradient-based search, tabu search methods, etc., very many times (eg, thousands of times, hundreds of thousands of times). Is based on the construction of a metamodel that requires iteration.
[0008]
However, industrially, it is usually difficult to obtain a large number of repeated experimental values necessary to use the metamodel, and further, it is inefficient to use the above method.
[0009]
The simplest method for finding the optimum buffer size is a method in which simulation is applied to all possible combinations of buffer sizes, and each execution result is evaluated. When this method is adopted, for example, assuming that each of three buffers can take a value of size 1 to 10, even if there are only three buffers, the total number of size combinations is 1,000. It also becomes.
[0010]
Since this method is not practical, in many cases, focus only on buffers that are empirically expected to be effective in improving the system, and apply simulations by discretely changing the size of such buffers. The method is adopted. However, this method cannot find a truly optimal size combination.
[0011]
Furthermore, as a technique for reducing the number of simulation trials, it is possible to use the “experiment design method”, but in any case, it is necessary to perform a large number of simulations in order to obtain the optimum combination of buffer sizes. This is not to be pointed out again.
[0012]
Optimizing the allocation of buffers to the system is the ultimate challenge, but to do so, the effect that the buffer has on each object in the system (including machines and other buffers), for example, changing the buffer ( It is necessary to define the relationship between the change position or change amount) and the reduction time of the stubbing time or blocking time of each object, and estimate the reduction amount of the staging time or blocking time from the change of the buffer.
[0013]
Furthermore, in order to accurately estimate the effect of such a buffer, the cause of the stave or block of each component of the system (also referred to as “each object”, for example, processing equipment in the production line) It is necessary to accurately estimate the buffer that is and the causal route that is the route in which the influence of the buffer reaches the stave or block of each object.
[0014]
Against the background described above, in the present invention, in order to cause the flow of the processed element from the upstream side to the downstream side, a plurality of components to be processed on the processed element are predetermined. For a system that is arranged according to the flow, the task is to estimate the cause element that causes each component to fall into an idle state and the cause path that causes the component to spread to each component It is.
[0015]
[Means for Solving the Problems]
The following aspects are obtained by the present invention. Each mode is divided into sections, each section is numbered, and is described in a form that cites other section numbers as necessary. This is to facilitate understanding of some of the technical features described herein and some of the combinations thereof. The technical features and combinations thereof described herein are It should not be construed as limited.
[0016]
Further, describing each section in the form of quoting the numbers of the other sections does not necessarily prevent the technical features described in each section from being separated from the technical features described in the other sections. It should not be construed as meaning, but it should be construed that the technical features described in each section can be appropriately made independent depending on their properties.
[0017]
(1) In order to generate a flow from the upstream side to the downstream side of the element to be processed, a plurality of components to be processed with respect to the element to be processed are arranged according to a predetermined flow. Cause / path for estimating a cause element that is another component that causes an idle state in a component and a cause path from the cause element to the idle component that is the component in the idle state An estimation method,
Based on the system data representing the change in the state of each component in the system over time, the relationship between the change in the state of the idle component and another component adjacent thereto is determined and determined. Based on the relationship, a first tracking step of tracking the idle component as a first starting element and another component adjacent thereto in a corresponding direction of the forward and reverse directions of the flow;
Based on the system data and according to a predetermined determination rule, by determining whether or not the traced component corresponds to the cause element, a cause identifying step for identifying the cause element,
If the causal element is not identified in the cause identifying step only by the execution result of the first tracking step, each of at least one tracking is finally traced based on the system data until it is identified. Determining a mutual relationship between each changed state state of the measured component and another adjacent component, and based on the determined relationship, the last tracked component is determined as the second starting component. A second tracking step performed with a content that tracks another component adjacent thereto in the direction corresponding to the forward direction or the reverse direction of the flow,
A path specifying step for identifying, as the causal path, a path that causes the causal element to affect the idle component in association with the causal element specified by the causal specifying process based on the system data;
Causal / route estimation method.
[0018]
If the flow of multiple components in a system is specified, and the flow of processed elements in the system is specified, the influence that any two adjacent components in the system can have on each other according to the rule of thumb It is possible to estimate several possibilities. This will be described in detail later with reference to FIGS.
[0019]
In such an environment, if system data representing the change in the status of each component in the system over time is used, the relationship between the change in the status over time of any component and another adjacent component can be obtained. Can be determined. The relationship is then used to select an appropriate one of the several possibilities described above, so that the relationship between two specific components adjacent to each other, eg, one component is idle. In the state, the other component constitutes the cause of the idle state, and the path between the two components constitutes a path that causes the cause to spread to the idle component It is possible to estimate it automatically.
[0020]
Furthermore, such estimation can be performed efficiently in a short time without depending on many iterations of simulation that simulates the behavior of the system in time series.
[0021]
Based on the knowledge described above, in the method according to this section, each of the idle component and another adjacent component is changed over time based on the system data representing the change in the state of each component in the system over time. The relationship between various state changes is determined.
[0022]
Further, based on the determined relationship, an idle component is used as a starting element, and another adjacent component is traced in a direction corresponding to the forward direction or the reverse direction of the flow of the processed element in the system.
[0023]
As a result, a cause element that is another component that causes an idle state in a certain component and a cause route that is a path from the cause element to the idle component are estimated.
[0024]
Therefore, according to this method, it is possible to estimate the causal element and the causal path without depending on many iterations of the simulation that simulates the behavior of the system in time series. The estimation can be performed in an extremely short time.
[0025]
By the way, in order to carry out the method according to this section, it is necessary to obtain the system data in advance, and this system data can be obtained by simulation that simulates the behavior of the system in time series. Is possible. In this case, a series of analyzes including acquisition of the system data and estimation of cause elements and cause paths (also referred to as “cause analysis” or “cause investigation”) must depend on simulation.
[0026]
However, when the method according to this section is implemented, the cause analysis does not have to depend on a large number of simulation iterations. Therefore, the series of analyzes described above is performed by a one-time simulation for acquiring system data. It can be completed.
[0027]
Therefore, according to this method, it becomes possible to perform a series of analyzes for the system efficiently and in a short time, and its practicality is improved.
[0028]
The “second tracking step” in this section may be performed only once or repeatedly several times. In the latter case, the “second starting element” in the current tracking matches the last tracked element in the previous tracking, and thus the second starting element is repeated each time the tracking is repeated. Updated to The same applies to the “second starting element” in the following items.
[0029]
(2) Further, when there is at least one cause element for a plurality of idle periods of the same idle component based on the system data, for each cause element, the effect on the idle component is expressed as follows: The cause / path estimation method according to (1), including an influence estimation step in which the influence is estimated in association with each cause path extending from each cause element to the idle component.
[0030]
In the system, there may be at least one cause element for a plurality of idle periods of the same idle component. In this case, according to the method according to this section, for each cause element, the influence of the cause element on the idle state of the idle component is estimated in relation to each cause route from the cause element to the idle component. Is done.
[0031]
Further, according to this method, the influence is estimated based on the system data.
[0032]
Therefore, according to this method, it is possible to estimate the influence individually for each cause element and in a short time without depending on many iterations of the simulation.
[0033]
(3) The first and second tracking steps are both upstream and downstream depending on the type of state of the corresponding one of the first and second starting elements prior to the tracking. The cause / path estimation method according to (1) or (2), including a direction setting step for setting the tracking direction to any one of the above.
[0034]
For example, as will be described in detail later, when the idle component of this time is in the starving state, there is a high possibility that the cause is upstream. Therefore, tracking for this idle component should then be done upstream.
[0035]
On the other hand, when the idle component of this time is in the blocking state, the cause is highly likely to exist on the downstream side. Therefore, tracking for the current idle component should then be done downstream.
[0036]
Based on such knowledge, in the method according to this section, the tracking direction is set in the direction corresponding to the type of the state of the starting element prior to tracking.
[0037]
Therefore, according to this method, tracking for specifying the cause / route can be easily performed in a short time without waste.
[0038]
(4) Both the first and second tracking steps are responsive to their respective states to all adjacent components adjacent to the corresponding one of the first and second starting elements prior to the tracking. A ranking step of selecting any adjacent component as a corresponding one of the first and second starting elements according to the assigned ranking,
The steps of (1) to (3), wherein the route specifying step includes a step of specifying each cause route between the idle component and each cause element based on the rank assigned to each cause element. The cause / route estimation method according to any one of the above.
[0039]
The cause of causing the idle component to fall into the idle state and the route thereof will be described in detail later, and can be classified into several patterns depending on the state of another component adjacent to the idle component.
[0040]
Based on this knowledge, in the method according to this section, prior to tracking, all adjacent components adjacent to the starting element are given ranks according to their respective states. Further, for each cause element, each cause route between the idle component and each cause element is specified based on the rank assigned thereto.
[0041]
(5) Further, during the tracking, based on the system data, for each buffer position between each component in the system, an amount by which the duration of the state of the idle component changes due to the change of the buffer; A statistical value calculation step of calculating a statistical value including at least one of a time during which each buffer continues to be the cause element and a time during which each adjacent component adjacent to the idle component remains on the cause path The cause / path estimation method according to any one of (1) to (4).
[0042]
The “buffer” in this section and the following sections can be interpreted to mean a buffer that exists in the system, or can be interpreted to mean a buffer that exists potentially. In the latter case, for example, a position where a buffer can actually exist is specified.
[0043]
(6) In order for the element to be processed to generate a flow from the upstream side to the downstream side, a plurality of components to be processed with respect to the element to be processed are arranged according to a predetermined flow. A system performance improvement amount estimation method for estimating an improvement amount for a system, in which the performance of the system is improved by changing each buffer between components of the system,
Based on system data representing a change in state of each component in the system over time, a first change amount in which the number of processed elements existing in the buffer changes for each component, and in the buffer A first estimation step of estimating a second change amount in which the number of spaces in which the element to be processed can be temporarily accommodated changes,
Based on the estimated first and second change amounts, for each component, a third change amount in which the number of processed elements that can be used by each component changes due to the change of the buffer; A second estimation step of estimating a fourth amount of change in which the number of spaces available to each component changes due to the change of the buffer;
A first idle time change amount in which the idle time of each component changes due to a change in the number of available processed elements for each component based on the estimated third and fourth change amounts And a third estimation step for estimating a total value of a second idle time change amount that changes due to a change in the number of available spaces,
A fourth estimation step for estimating the improvement amount by summing the estimated total values for all the components;
System performance improvement amount estimation method including
[0044]
In this method, (a) the first change amount in which the number of processed elements existing in the buffer changes, and (b) the number of spaces that can temporarily store the processed elements in the buffer change. A second variation, (c) a third variation in which the number of processed elements available to each component changes due to the buffer change, and (d) a number of spaces available to each component. Based on the fourth change amount that changes due to the buffer change, an improvement amount of the system performance due to the buffer change is estimated.
[0045]
Further, in the method according to this section, the first to fourth change amounts to be used for estimating the improvement amount are performed based on the system data.
[0046]
Therefore, according to this method, it is possible to estimate the amount of improvement in the system performance caused by the buffer change without depending on a large number of simulation iterations. Therefore, the estimation can be performed efficiently in a short time. It becomes easy.
[0047]
Here, the “first change amount” can be defined as an amount by which the number of elements to be processed existing in the buffer changes due to, for example, a change in the buffer. “Change amount” can be defined as, for example, an amount by which the number of spaces that can temporarily accommodate the processing target element in the buffer changes due to a change in the buffer.
[0048]
(7) The system performance improvement amount estimation method according to (6), wherein the second estimation step estimates the third and fourth change amounts based on the statistical value in the item (5).
[0049]
(8) The first estimation step includes
Based on the representative bottleneck probabilities for all components upstream of each buffer and the representative bottleneck probabilities for all components downstream of the plurality of components, the buffers The system performance improvement amount estimation method according to item (6) or (7), wherein at least one of a full rate and a depletion rate is estimated, and the first and second change amounts are estimated based on the estimated value.
[0050]
The “bottleneck probability” in this section can be defined, for example, as the ratio of the length of time each component in the system is active to the total time (eg, total analysis time for the system) . The bottleneck probability defined in this way means, for example, a bottleneck period occupancy rate in which a bottleneck period in which each component in the system functions as a bottleneck occupies the entire time. An example of a method for calculating the bottleneck period occupation rate is described in Japanese Patent Application No. 2002-81077 of the present applicant.
[0051]
(9) A program executed by a computer to implement the method according to any one of (1) to (8).
[0052]
If this program is executed by a computer, the same effect can be realized according to basically the same principle as in the method described in any one of the above items (1) to (8).
[0053]
This program can be interpreted to include not only a combination of instructions executed by a computer to fulfill its function, but also files and data processed by the computer according to each instruction.
[0054]
(10) A recording medium on which the program according to item (9) is recorded so as to be readable by a computer.
[0055]
If the program recorded on the recording medium is executed by a computer, basically the same effect can be realized according to the same principle as in the method described in any one of the above items (1) to (8).
[0056]
The “recording medium” in this section can adopt various formats, for example, a magnetic recording medium such as a flexible disk, an optical recording medium such as a CD and a CD-ROM, a magneto-optical recording medium such as an MO, and an ROM. At least one of removable storage and the like can be employed.
[0057]
(11) A system analysis apparatus that performs the method according to any one of (1) to (8) using a computer.
[0058]
According to this apparatus, the same effect can be realized in accordance with basically the same principle as in the method described in any one of the items (1) to (8).
[0059]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, one of more specific embodiments of the present invention will be described in detail with reference to the drawings.
[0060]
A buffer prediction method according to an embodiment of the present invention will be described briefly. As illustrated conceptually in a flowchart in FIG. 1, a discrete event system having a plurality of components (hereinafter simply referred to as “system”). 2 includes a cause / path estimation step of estimating a cause element that is a component that constitutes a cause of causing each component element to be in an idle state and a cause path that causes the cause to spread to each component element. This buffer prediction method can be restated as a buffer laying effect prediction method.
[0061]
This buffer prediction method further includes a buffer change (change position or change amount) in the system and a staging time (continuation time of parts shortage state) or blocking time (continuation time of parts unloadable state) of each component. It includes a reduced starving / blocking time estimation step that estimates the amount of reduction in starving time or blocking time from buffer changes according to a time independent mathematical relationship with the amount of reduction.
[0062]
The buffer prediction method further includes a buffer optimization step of optimizing the buffer allocation (with respect to installation position and buffer capacity) to the system.
[0063]
The system is composed of a plurality of elements that perform various actions. This system does or does not contain the object to be processed that does not act. The element can act on the object to be processed by, for example, processing (deforming) or conveying the object to be processed. Elements can also act on other elements.
[0064]
An example of such a system is for a production line. The object to be processed there is a part to be processed, that is, combined, deformed or separated. The element that performs the process is a machine that performs the process.
[0065]
However, the elements are not only those that deform the object to be processed. There may also be elements that carry the object to be processed, such as automated guided vehicles or workers. Some elements can be serviced against other elements, for example, as a repair team that repairs other elements when necessary. Furthermore, other elements can be used to carry the object to be processed into the system and to remove the object to be processed from the system. Finally, some elements can be utilized for some combinations of the above mentioned applications.
[0066]
Although the elements described above are general, the present invention can be applied to other elements not described here.
[0067]
Usually all of these elements are limited by the number of actions that can be performed in a certain amount of time. Thus, the entire system is limited by the number of parts that can pass through the system in a certain amount of time. Some factors have a greater impact on the overall system performance than others. In this embodiment, the effect of various factors on the overall system performance is measured. In many cases, some factors are more than others and limit the overall system throughput (throughput). These elements are commonly referred to as bottlenecks or constraints.
[0068]
In this embodiment, the operating system is observed to obtain the necessary information. This observation is performed, for example, by looking at the production line or collecting simulation data. The necessary information is a list of data (time-series data) for each element, and indicates the state of each element in each period in detail.
[0069]
FIG. 2 shows an example of the state of a machine (for example, a processing machine) during a certain period of time. The machine operates on a part for some period, waits for a new part as an example of a staging state, which is an example of an idle state in another period, and waits on the unloading side in another period. The queue is full because the queue is full, waits for all processed parts to be ejected, and fails for another period to be repaired.
[0070]
Here, the blocking state means that, for example, a device following the machine, such as a buffer, another machine, or a conveyor belt, cannot handle the part in an operating state, so that the machine carries out the finished part. It means a state that can not be.
[0071]
FIG. 3 shows an example of a data set for various machines. In this example, a machine name and time are given for each change in machine state. In the figure, the state 0 is the operating state, the state 1 is the staging state, the state 2 is the blocking state, and the state 3 is a state where a failure occurs and self-repair is being performed. The list of data shown in the figure is necessary for carrying out this embodiment.
[0072]
The theoretical background and specific method of this buffer prediction method will be described.
1 Simulation example
[0073]
FIG. 4 conceptually shows a block diagram of an example of a system in which this buffer prediction method is implemented. This buffer prediction method is implemented based on the temporal behavior of the system analyzed by a single computer simulation (or obtained by measuring a real system). Therefore, the example illustrated in FIG. 4 is also an example of a simulation target.
[0074]
Here, the term “simulation” is, for example, “a process of simulating an entity and changing a model expressed in a computer program language in time series, thereby acquiring information on the dynamic behavior of the entity”. Can be defined as In this way, the simulation can be considered as a time-dependent process.
[0075]
The above system example is a production system, and includes seven machines with names M1 to M7 as shown in FIG. These machines flow along a production line constituted by an upstream straight portion, a branch portion branched into two flows, and a downstream straight portion.
[0076]
Furthermore, in this production system, nine buffer positions are set in front of or behind each machine. The buffers installed in front of the machines M3, M5 and M6 are labeled BM3, BM5 and BM6, respectively, whereas they are installed behind the machines M1, M2, M3, M4, M5 and M6. The buffers are labeled AM1, AM2, AM3, AM4, AM5 and AM6, respectively. In this production system, the first machine M1 is never starved and the last machine M7 is never blocked.
[0077]
In this production system, the cycle time of the machine is constant, but the machine is irregularly delayed by exponentially distributed failure times. When a simulation was performed on this production system using a certain simulation software, the time was 500 days for the simulation.
[0078]
In this production system, the machine M3 is a main bottleneck, and the machines M2, M5, and M6 are secondary bottlenecks, and the system is well balanced. In that system, one part is completed every 55.7 s (seconds), that is, an average of 64.7 parts are completed per hour.
[0079]
2 Buffer prediction model
[0080]
This buffer prediction method is implemented assuming a buffer prediction model. This buffer prediction model is constructed for the purpose of estimating the effect on the processing capacity of the system due to the increase in the size of the buffer.
[0081]
2.1 Overview of buffer prediction model
[0082]
A method for predicting the effect of changing the buffer size in the production system will be briefly described briefly. FIG. 5 is a flowchart showing an outline of the buffer prediction model, and further describes some symbols used in the buffer prediction method. 69 and 70 show definitions of these symbols.
[0083]
Prior to implementing this buffer prediction method, it is necessary to perform statistical analysis on the initial system many times. This analysis includes (a) basic statistical analysis to determine machine availability, initial system performance, and machine latency (also referred to as “idle time”) distribution; b) There is bottleneck detection analysis for determining the bottleneck probability that each component of the system acts as a bottleneck.
[0084]
The buffer prediction method further includes a cause factor of machine staging and blocking, a block / starve (meaning “block or starve”, and the same shall apply hereinafter) and a cause / factor of the block / starve. In order to identify the buffer in between, it is necessary to analyze the events of starving and blocking.
[0085]
After completion of this initial statistical analysis, the buffer capacity can be selected and changed. This can be done by an analyst or automatically by an optimization program. Based on the probability of bottlenecks and other information, the average number of parts and average number of spaces in the modified buffer are estimated.
[0086]
There are many methods that can be used to make this estimation. In this embodiment, the average added number of parts / spaces in all buffers (meaning “parts or spaces”; the same applies hereinafter) and Based on the cause of machine blocking and stubbing, for each machine i, an additional amount of average parts ΔFi and an additional amount of average empty space ΔEi are estimated.
[0087]
In the next step, the time proportion at which idle time is reduced is analyzed for all machines. This process is somewhat complicated and requires a lot of additional data, such as machine waiting time distribution, operating rate, unit action time (Time Per Part) which is the action time per part. The estimated decrease in machine idle (blocked or starved) time is then summarized into an estimate of the decrease in system performance by using the bottleneck probability. This ultimately makes it possible to predict the overall system performance of the modified system.
[0088]
2.2 Causes of starving or blocking
[0089]
The buffer is intended to reduce machine blocking and staging periods and to facilitate the flow of objects in the system. Then, in the first step, the cause of the blocked or starved machine is determined.
[0090]
To know the effectiveness of the buffer, it is necessary to know which machine is starved or blocked by insufficient buffer capacity. Thus, all blocking and staging events are followed to determine the blocking and staging path as well as the elements that make up the cause of the block or stave. Before detailing the overall buffer prediction method, we first introduce a list of some case studies.
[0091]
2.2.1 Case study
[0092]
2.2.1.1 Straight line
[0093]
The simplest case is a straight line. FIG. 6 shows a simple line with three machines M1, M2, M3 and two buffers B1, B2.
[0094]
Obviously, the block in machine M2 is due to buffer B2. However, the buffer B2 is not necessarily the original cause of the block. If buffer B2 is full, the block is caused by the last machine M3 on the same line. Thus, the block in machine M2 is caused by machine M3 via buffer B2.
[0095]
However, there is also the possibility that buffer B2 is not full and only the speed of that buffer B2 causes a slight delay of machine M2. In this case, the block is attributed to buffer B2.
[0096]
Similar logic can be used to analyze the staging of machine M2. When machine M2 is starved, it will be attributed to buffer B1. If buffer B1 is empty, the stave is attributed to the first machine M1. If the buffer B1 is not empty, the stave may be attributed to that buffer B1 due to the insufficient speed of the buffer B1.
[0097]
2.2.1.2 Branched line
[0098]
If the line is branched, the overall approach becomes more complex. FIG. 7 shows a branching system with four machines. No buffer is set for simplicity. If machine M1 is blocked, the block may occur via machine M2 or machine M3, or both, and is due to any of the following three machines M2, machine M3, and machine M4. there's a possibility that.
[0099]
If blocking is due to machine M4, it is important not only to determine the causal element of the block, but also to determine the path of the block. For example, if the machine M1 is blocked by the machine M4 via the machine M2, the machine M3 is not related to the block of the machine M1, so the improvement of the buffer around the machine M3 is not necessarily the processing capacity of the line. It will not improve.
[0100]
2.2.1.3 Reverse blocking
[0101]
For more complex cases, a machine may be blocked by another machine, which may be in parallel or upstream rather than downstream from the machine.
[0102]
In the example of FIG. 8, machine M1 and machine M2 supply parts for machine M3, and machine M3 requires parts from both machine M1 and machine M2 to work (eg, Machine M3 assembles parts from machine M1 and parts from machine M2).
[0103]
If machine M1 is blocked, machine M3 cannot receive parts from machine M1. This may be because the machine M3 is in operation. However, machine M3 may be starved because it cannot receive parts from machine M2. Thus, the block in machine M1 is attributed to machine M2, that is, the cause of that block may be traced downstream from machine M1 to machine M3 and then back to machine M2 upstream. .
[0104]
2.2.1.4 Reverse staging
[0105]
As shown in FIG. 9, a similar case is observed even in the stubbing state. Machine M2 may be starved because machine M1 is blocked by machine M3. Therefore, the stave of machine M2 is attributed to machine M3.
[0106]
2.2.2 General theory of blocking and staging
[0107]
The logic for tracing the events of starving and blocking upstream or downstream will be described in detail. While that logic applies to most cases, there are situations where the exact cause of a block or stave is sometimes not determined. The logic described below allows the best estimation, but sometimes requires more information to determine the exact cause. Also, if multiple upstream or downstream machines or buffers are available, a ranking is performed to determine the most probable cause of blocking or staging.
[0108]
2.2.2.1 General theory of stubbing
[0109]
There are basically six possibilities for upstream objects. FIG. 10 shows an explanation of the notation of FIG. 11 and FIG.
[0110]
Assuming that one machine has been starved, there is one object (buffer or machine) upstream of the starved machine. FIG. 11 shows five possible objects. These objects are as follows.
[0111]
Starving situation (1)
[0112]
The upstream object is also a starved machine. In this case, the cause investigation of stave must continue upstream.
[0113]
Starving situation (2)
[0114]
The upstream object is an empty buffer. In this case, the staging cause investigation must continue upstream by determining the cause of the empty buffer.
[0115]
Starving situation (3)
[0116]
An upstream object is a machine in an active state. In this case, starving is caused by the machine in its working state. In this case, the investigation of the cause of starving is completed.
[0117]
Starving situation (4)
[0118]
The upstream object is a blocked machine. In this case, stubbing is caused by the blocked upstream machine, and therefore the blocked machine must be followed downstream to determine the ultimate cause of that staging.
[0119]
Starving situation (5)
[0120]
The upstream object is a buffer that is neither empty nor full. In this case, it is never clear, but it is in the process of shipping the part to the machine where the buffer is starved, but because of its insufficient buffer speed, the buffer (not empty) has caused the downstream machine to star. Assume. In this case, the staging cause investigation is completed.
[0121]
2.2.2.2 General theory of blocking
[0122]
Similar to the case of single machine staging by an upstream machine, a single machine can be blocked by a downstream object. FIG. 12 shows five basic possibilities for the blocking situation.
[0123]
Blocking situation (1)
[0124]
The downstream object is also a blocked machine. In this case, the investigation of the cause of blocking must continue downstream.
[0125]
Blocking situation (2)
[0126]
The downstream object is a full buffer. In this case, the cause of blocking must be continued downstream by determining the cause of the full buffer.
[0127]
Blocking status (3)
[0128]
A downstream object is a machine in an active state. In this case, blocking is caused by the machine in action. In this case, the cause investigation of the blocking is completed.
[0129]
Blocking status (4)
[0130]
The downstream object is a starved machine. In this case, blocking is caused by the starved downstream machine, and therefore the starved machine must be traced upstream to determine the ultimate cause of blocking.
[0131]
Blocking status (5)
[0132]
The downstream object is a buffer that is neither empty nor full. In this case, it is never clear, but the buffer is in the process of receiving parts from the blocked machine, but due to its insufficient buffer speed, the buffer (which is never full) will block the upstream machine. Assuming that In this case, the cause investigation of the blocking is completed.
[0133]
2.2.2.3 Ranking of starving
[0134]
A single machine does not have only one upstream or downstream object as a causal factor for its staging, but a plurality of objects on the upstream and downstream sides (for example, one A plurality of facilities having a parallel relationship with each other). In this case, an algorithm must be used to investigate the cause of starving.
[0135]
For the investigation, a plurality of machines on the upstream side and the downstream side are ranked. The weighting method is used when one or more identical objects exist on the upstream side. For example, if there are two machines that stir a machine upstream of it, both of which are in action, it is determined that the two machines have half the cause of the stave. The Ranking is performed for a plurality of objects upstream of the machine being starved as follows.
[0136]
Rank 1: Starving situation (1)
[0137]
If another starved machine is present upstream, follow the starved machine upstream. When there are a plurality of machines that have been starved on the upstream side, each of them is traced and the same weight is given to those paths.
[0138]
Rank 2: Starving situation (2)
[0139]
If an empty buffer exists on the upstream side, the empty buffer is traced upstream. Again, if there are a plurality of empty buffers on the upstream side, each of them is traced and the same weight is given to those paths.
[0140]
Rank 3: Starving status (3)
[0141]
If there is a machine in action upstream, stubbing is caused by the machine in action. When there are a plurality of machines in the active state on the upstream side, they follow each of them and give the same weight to their routes.
[0142]
Rank 4: Starving status (4)
[0143]
If a blocked machine is present upstream, the blocked machine is traced downstream to investigate the cause of blocking, which is also the cause of starving. When there are a plurality of blocked machines on the upstream side, each of them is traced and the same weight is given to those paths.
[0144]
Rank 5: Starving situation (5)
[0145]
If a non-empty buffer is present upstream, the machine in question is starved due to the speed of the upstream buffer. When there are a plurality of non-empty buffers on the upstream side, each of them is traced and the same weight is given to those paths.
[0146]
Thus, for example, if the active machine and the starved machine are upstream, the algorithm ignores the active machine (rank 3) and the starved machine (rank 1) upstream. Trace towards If two objects have the same rank, the exact cause cannot be determined, but both objects will be traced and there may be two causative factors of starving.
[0147]
2.2.2.4 Ranking blocking
[0148]
Similar to the starving situation, each object is ranked in the blocking situation. When there are a plurality of objects on the downstream side, the ranking is performed as follows.
[0149]
Rank 1: Blocking status (1)
[0150]
If there is another blocked machine downstream, follow the blocked machine downstream. When there are a plurality of blocked machines on the downstream side, each of them is traced and the same weight is given to the paths.
[0151]
Rank 2: blocking status (2)
[0152]
When an empty buffer exists on the downstream side, the empty buffer is traced downstream. When there are a plurality of empty buffers downstream, each of them is traced and the same weight is given to those paths.
[0153]
Rank 3: Blocking status (3)
[0154]
If there is a machine in action downstream, the blocking is caused by the machine in action. When there are a plurality of machines in the operating state on the downstream side, they follow each of them and give the same weight to their paths.
[0155]
Rank 4: Blocking status (4)
[0156]
If a starved machine is present downstream, the starved machine is traced upstream to investigate the cause of stubbing, which is also the cause of blocking. When there are a plurality of machines that are starved on the downstream side, each of them is traced and the same weight is given to the paths.
[0157]
Rank 5: Blocking status (5)
[0158]
If a non-empty buffer is present downstream, the machine in question is blocked due to the speed of the downstream buffer. When there are a plurality of non-empty buffers on the downstream side, each of them is traced and the same weight is given to those paths.
[0159]
2.2.3 Causes of blocking and stubbing
[0160]
The production system we are focusing on this time was analyzed to determine the cause of blocking and staging of all machines. FIG. 13 shows the probability that each machine is starved by another machine.
[0161]
FIG. 13 shows the percentage of time in which one machine (on the coordinate axis at the right side of the figure) was starved by another machine (on the coordinate axis at the bottom of the figure). The ratio of occupying the analysis time) is shown as a percentage. For example, noting the maximum stirability (highest pillar), machine M4 was starved by machine M3 at a time rate of about 26 percent. This is not surprising since machine M3 is the bottleneck of the system and machine M4 is much faster than machine M3 in this production system.
[0162]
Even more interesting is the staging of machine M3. Machine M3 was the main bottleneck of the system and was starved by the upstream machine M2 at a rate of approximately 2 percent. However, machine M3 was also starved by machine M5 at a rate of about 2.1 percent. However, the machine M5 does not exist upstream or downstream of the machine M3. Instead, machine M5 is on a path in parallel with machine M3. This means that machine M3 was starved by machine M2, which was blocked by machine M5. In some cases, machine M5 blocked machine M2, causing machine M3 to be starved by machine M6 and even machine M7 (at 0.8 percent and 0.7 percent time rates, respectively). .
[0163]
Similarly, machine M5 is not only starved by upstream machine M2 (at a rate of 3.2 percent) but also starred by machine M3 in parallel (at a rate of 1.9 percent). It was.
[0164]
FIG. 14 shows the probability that each machine is blocked by another machine. For example, machine M1 is blocked by a subsequent machine M2 (31 percent time percentage) and machine M6 is blocked by a downstream machine M7 (27 percent time percentage) for most of the total analysis time. . In this simulation, blocking due to stubbing in parallel paths does not occur.
[0165]
2.2.4 Blocking and staging paths
[0166]
The behavior of blocking and starving can also be expressed as shown in FIGS.
[0167]
FIG. 15 shows a path for generating blocking and a path for generating staring for the machine M3. Of all staging times, about 30 percent is attributed to machine M2, and about 60 percent is attributed to or through machine M5. Machine M5 generates 30 percent of the staging time of machine M3, and machine M6 and machine M7 each generate 15 percent of stubbing time by passing through machine M5.
[0168]
Machine M3 is further blocked from time to time. All blocking passes through machine M4, but there is little blocking due to machine M4. Instead, machine M3 is mostly blocked by machine M7 and further blocked by machine M6.
[0169]
In FIG. 16, a similar graph is created for the machine M5. Machine M5 is not only stared by upstream machine M2, but is also starved by machine M3 in parallel with it and is also starved by machine M4. Downstream blocking is due to machine M6 and machine M7 only, as expected.
[0170]
2.2.5 Briefing on blocking and staging
[0171]
The increase in buffer reduces the blocking and staging of associated machines. The above section details which buffers potentially affect machine blocking and staging. Theoretically, a large number of buffer positions can be taken. For simplicity of explanation, the algorithm assumes that buffers can be placed immediately before and after each machine. There are a total of four modes in which one buffer relaxes another machine. These four modes are listed next and are further tabulated in FIG.
[0172]
Mode 1
[0173]
Increasing the capacity of the empty buffer reduces the blocking of another machine (eg, in FIG. 15, increasing the buffer capacity in buffer AM4 reduces the blocking of machine M3).
[0174]
Mode 2
[0175]
Increasing the empty buffer capacity decreases the staging of another machine (eg, in FIG. 16, increasing the buffer capacity in buffer BM3 decreases the staging of machine M5).
[0176]
Mode 3
[0177]
Increasing the number of parts in the buffer reduces the blocking of another machine (eg, in FIG. 8, increasing the number of parts in the buffer AM1 behind machine M1 reduces the blocking of machine M2). .
[0178]
Mode 4
[0179]
Increasing the number of parts in the buffer decreases the staging of another machine (eg, in FIG. 15, increasing the number of parts in the buffer BM3 decreases the staging of the machine M3).
[0180]
Thus, when analyzing the blocking and staging situations, the percentage of time that one buffer position affects another machine is determined. The time ratios are shown in tables in FIGS. 18 to 21 for the above-described modes 1 to 4, respectively. The results shown in FIG. 18 to FIG. 21 are obtained by the computer 20 executing the routine shown in FIG.
[0181]
FIG. 18 shows the time ratio (the ratio of the blocked time to the total time. Hereinafter, also simply referred to as “time ratio”) for each mode. May be reduced if more free space is available in the buffer.
[0182]
For example, machine M1 is blocked at a time rate of 46.1 percent. Each time the machine M1 is blocked, the block passes through its adjacent buffer AM1. If free space available for this buffer AM1 is added, the blocking of the machine M1 is reduced.
[0183]
However, machine M1 is blocked only at a time percentage of 4.4 percent, and that block passes through buffer BM3. By adding available space to the buffer BM3, it is possible to reduce only 4.4% of the unloading time interval TBP of the machine M1 by reducing the blocking time. Let PBEBj, i represent the percentage of time that the empty space in buffer j reduces the blocking of machine i.
[0184]
In addition, the unloading time interval TBP means the time interval between the moment when one preceding processed component is unloaded from a certain machine and the moment when one subsequent processed component is unloaded. Also called unloading part interval. However, the unloading time interval for each machine does not necessarily match the unloading time interval for the entire system.
[0185]
FIG. 19 shows, for mode 2, the percentage of time each machine is starved, which can be reduced if more free space in the buffer becomes available. This phenomenon sometimes occurs in a branching system, where one machine is starved because another machine is blocked and another machine's starving is reduced if the blocking of another machine is reduced. Decrease.
[0186]
For example, because machine M2 is blocked by buffer BM5, machine M3 is starred at a time rate of 3.4 percent. The effect that the empty space in buffer j has on the stave of machine i is called PBESj, i.
[0187]
FIG. 20 shows the percentage of time each machine is blocked for mode 3, with the block decreasing as more parts are available in another buffer. This phenomenon sometimes occurs in a branching system, where one machine is blocked because another machine is starved, and when another machine's starve is reduced, that machine's Block is reduced.
[0188]
However, the described example does not show such an effect, and each value in the table of FIG. The effect that the increase in the number of parts in the buffer j has on the blocking of the machine i is represented by PBFBj, i.
[0189]
FIG. 21 shows the percentage of time that each machine is starved for mode 4, and that block may decrease as more parts are available in another buffer. For example, machine M6 is starved at a rate of 6.5 percent through buffer AM4. The effect that the increase in the number of parts in the buffer j has on the staging of the machine i is referred to as PBFSj, i.
[0190]
3.3 Effective buffer size
[0191]
The above describes how an increase in the number of parts or the capacity of the buffer affects the idle time of another machine. However, in the buffer, the number of parts or available capacity is not infinite. First, most buffers are limited by size. Secondly, the number of parts or available buffer space is actually dependent on the average number of parts in the buffer and the size of the buffer.
[0192]
For example, in the production system shown in FIG. 4, the buffer AM1 affects the blocking of the machine M1. Adding free space in the buffer AM1 reduces the possibility of the machine M1 being blocked. However, the buffer AM1 is almost always full. Even if the buffer space is increased by a small amount, the space is almost immediately filled, and the machine M1 is not changed when the long-time behavior is observed.
[0193]
Therefore, in order to estimate the average number of free spaces and the average number of available parts, it is necessary to determine the average number of parts in the buffer. For example, if the buffer AM3 has a size of 10 and has an average number of parts of 4, this means that on average 4 parts are available for the machine M4, On average 6 spaces (subtracting 4 parts from 10 capacities) mean that the machine M3 is available for unloading parts.
[0194]
As shown by an equation in FIG. 22, an average full rate at which a certain buffer j is full is indicated as PBFj, and the relationship between the average full number of parts BFj in the buffer is configured using the buffer capacity BCj. According to this equation, the average number of parts BFj is calculated for the buffer j by the product of the average fullness rate PBFj and the buffer capacity BCj. This equation also shows the relationship between the average depletion rate PBEj in which the buffer j is empty and the average number of free spaces BEj in the buffer. According to this equation, the average free space number BEj is calculated by the product of the average depletion rate PBEj and the buffer capacity BCj.
[0195]
In this equation, the relationship between the average fullness rate BFj and the average depletion rate PBEj and the relationship between the average number of parts BFj, the average number of spaces BEj, and the buffer capacity BCj are also shown.
[0196]
In addition, some mathematical expressions used in the present embodiment include those not using time as a variable and those using time as a variable. When a mathematical expression that does not use time as a variable is used, a solution can naturally be obtained without performing a simulation that simulates the system in time series. In contrast, in this embodiment, a mathematical expression using time as a variable expresses a statistical rule. Therefore, this mathematical expression is also solved without performing a simulation that simulates the system in time series. Can ask for.
[0197]
Therefore, in the present embodiment, all necessary solutions can be obtained without such a simulation, and the computer calculation time for system analysis can be greatly reduced.
[0198]
3.3.1 Experimental analysis
[0199]
There are two basic approaches to determine the average number of parts in the buffer. The simplest approach is simply to measure the average number of parts in the buffer. This is the simplest method of measuring to determine the average number of parts in a buffer, assuming that a measurable buffer exists in the simulation. This technique does not work when the buffer size is zero.
[0200]
Furthermore, in some systems, the average number of parts in the buffer may change as the buffer size increases. For example, FIG. 23 shows a histogram of the buffer AM3 when the buffer size is 1. The average number of parts in the buffer is 0.05 parts, or 5 volume percent (5 percent of the total volume).
[0201]
However, when the buffer size changes to 10, these values change. FIG. 24 shows a buffer with a size of 10 as a histogram. In this example, there are no more than six parts at any time. Thus, when the buffer size changes to 10, the average number of parts changes to 0.14 or 1.4 volume percent. Therefore, both the average number of parts and the capacity change. Therefore, when the buffer size is increased, it becomes more difficult to predict the number of available parts.
[0202]
3.3.2 Theoretical analysis
[0203]
The problem in experimentally analyzing the average part average in the buffer is that the buffer must be in the simulation. If there is no buffer, the average number of parts cannot be measured. It is therefore interesting to theoretically estimate the average number of parts in a buffer without actually measuring the actual buffer.
[0204]
In this buffer prediction method, the average number of parts in the buffer is estimated without actual buffer measurement values. This estimate is based on the bottleneck probability that a machine is a bottleneck in a system. The basic concept is that buffers upstream of the bottleneck tend to be full, while buffers downstream of the bottleneck tend to be empty.
[0205]
The average number of parts in the buffer can be estimated from the relationship between the bottleneck probabilities between the upstream side and the downstream side. The example used here is the example shown in FIG. FIG. 25 shows the bottleneck probability for an initial system with almost all buffer capacities of 0 or 1 (total bottleneck probability, ie, the probability that a machine is a bottleneck alone and The sum of the probability that the bottleneck is a shifting bottleneck that shifts) is shown in the table.
[0206]
Assume the following example for buffer AM1. The first step in the present technique is to calculate the total value of the bottleneck probabilities for the upstream side and the downstream side of the buffer AM1.
[0207]
The upstream bottleneck probability is simply the bottleneck probability of machine M1, which is 0 percent. Therefore, the total value of the upstream bottleneck probability is 0 percent.
[0208]
As shown in FIG. 26, the downstream bottleneck is slightly more complicated than the upstream bottleneck because the current system branches behind the machine M2. Currently, the branching part is combined with a technique of reducing the weight of the branched bottleneck to ½ and tracing each branch to the end. Therefore, the total value of the downstream bottleneck probabilities is equal to the bottleneck probability of the machine M2 and 1/2 the bottleneck probability (weighted bottleneck probability) for the first branch M3-M4-M6-M7. It is the sum of the bottleneck probability ½ (weighted bottleneck probability) for the second branch M5-M6-M7.
[0209]
The bottleneck probabilities for machine M6 and machine M7 are weighted as ½ at their respective branches, so they are used to calculate the total bottleneck probabilities for these machines M6 and M7. Therefore, the total value of the downstream bottleneck probabilities (total weighted bottleneck probability) of the buffer AM1 is M2 + (M3 + M4 + M6 + M7) / 2 + (M5 + M6 + M7) / 2 = 121 percent. This total value can exceed 100 percent because it includes both shifting bottleneck effects and single bottleneck effects.
[0210]
Therefore, the fullness rate Z of the buffer AM1 is obtained by setting the downstream bottleneck probability Y (the sum of the bottleneck probability of M2 and the weighted bottleneck probability for (M3 + M4 + M6 + M7) and the weighted bottleneck probability for (M5 + M6 + M7)) to the upstream bottle. It is estimated as the result of dividing by the total value of the neck probability X (bottleneck probability of M1) and the downstream bottleneck probability Y. The relationship between those three is
Z = Y / (X + Y)
It is represented by Thus, in this example, buffer AM1 is estimated to be nearly 100 percent full.
[0211]
On the other hand, the fullness ratio Z of the buffer BM1 is estimated as the upstream bottleneck probability X divided by the total value of the upstream bottleneck probability X and the downstream bottleneck probability Y. The relationship between those three is
Z = X / (X + Y)
It is represented by
[0212]
In FIG. 27, the estimated value (predicted value) is shown in a table with respect to the buffer fullness rate while being compared with the measurement result (actual value). The measurement result is caused by the configuration of various types of buffers, and does not include the effect that the change of the system has on the bottleneck probability and thus on the estimated value of the buffer. For some buffers, no measurements have been taken.
[0213]
In FIG. 28, the relationship between the buffer size and the average number of parts for a plurality of buffers is shown in a plurality of graphs in comparison with the measured value and the predicted value. However, the scales of these graphs are not all the same.
[0214]
The measured values and predicted values shown in FIG. 28 are based on the bottleneck probability of the initial system (the system before the buffer capacity is increased). The prediction model from which these prediction values are derived is based on the simulation default settings (buffers AM1, AM2, AM3, BM6 and AM6 have a capacity of 0 and buffers BM3, AM4, BM5 and AM5 have a capacity of 1). ing.
[0215]
It can be seen that this prediction model is appropriate for most buffers. The predicted values for buffers AM2, BM3, BM5, and AM5 are accurate enough to be used in a model that analyzes the buffer throughput.
[0216]
3.3.2.1 Prediction error due to change in bottleneck probability
[0217]
When the buffer size changes greatly, the bottleneck probability changes. Since the average number of parts in the buffer is based on the bottleneck probability, the predicted value of the average number of parts in the buffer also changes.
[0218]
FIG. 29 shows a table of bottleneck probabilities (overall bottleneck probabilities, that is, total values of single bottleneck probabilities and shifting bottleneck probabilities) for two types of buffer configurations. In the initial buffer (buffers AM1, AM2, AM3, BM6 and AM6 have a capacity of 0 and buffers BM3, AM4, BM5 and AM5 have a capacity of 1), machine M3 has a bottleneck probability of 62 percent. Machine M2, Machine M5 and Machine M7 are also important secondary bottlenecks with bottleneck probabilities in the range of 20 percent and 30 percent.
[0219]
The rightmost column in FIG. 29 shows the bottleneck probabilities for the buffers AM2, BM3, AM3, AM4, BM5, and AM5 whose capacity has been increased to 20. In this case, clearly, machine M3 is the main and only bottleneck with a 99% bottleneck probability and the remaining machines are less important bottlenecks with a bottleneck probability of 2% or less.
[0220]
FIG. 30 tabulates the buffer fullness measurements and predictions for both the initial system and the system with the buffer increased in capacity. As the bottleneck probability has changed, the buffer fullness has also changed.
[0221]
FIG. 28 is compared with FIG. FIG. 31 also shows a comparison between the predicted value of the buffer fullness and the measured value. However, the predicted value is based on the bottleneck probability for the buffers AM2, BM3, AM3, AM4, BM5 and AM5 whose capacity has been increased to 20 (see the rightmost column in FIG. 29). Buffers AM2, BM3, BM5 and AM5 are fairly accurate, and buffers AM3 and AM4 are also close to true measurements.
[0222]
3.3.2.2 Prediction error due to decrease in average number of parts in buffer
[0223]
In the above prediction model, the average number of components in the buffer is given as a percentage of the buffer capacity (ratio of the number of components in the buffer). Therefore, calculation and comparison can be easily performed between different buffers.
[0224]
However, the percentage of parts in the buffer can vary as a function of buffer capacity. For example, in a simple queuing system with only one machine that has a machine capacity that exceeds the supply, the average number of parts in the buffer can be calculated. For example, if the machine can process 6 parts per hour (exponential distribution) and 5 parts per hour (exponential distribution) arrive, the average number of parts in the buffer is 4.2 Can be calculated as For large buffers (capacity> 15), the average number of parts in the buffer is always about 4.2.
[0225]
Accordingly, the ratio of the number of components in the buffer decreases as the buffer capacity increases and the number of components is maintained. Similarly, for small buffers, the in-buffer component ratio may increase due to the waiting time distribution in the queue.
[0226]
A description will be given by taking the buffer AM4 at the rear of the machine M4 as an example. Since this buffer has little effect on the performance of the system, this example is a convenient example for studying component counts for various buffer sizes without being interfered by changing bottlenecks.
[0227]
FIG. 32 shows the buffer as a histogram when the capacity is 5. The buffer rarely accommodates more than two parts and has no time to accommodate more than three parts. Therefore, buffer sizes greater than 3 remain unused.
[0228]
This is illustrated graphically in FIG. 33, where the average number of parts in the buffer for various buffer sizes is shown on the left scale. For buffer sizes greater than three, the average number of parts hardly changes. As the number of components is maintained and the buffer size increases, the ratio of the number of components in the buffer decreases. This is shown on the right scale in FIG.
[0229]
Therefore, the predicted value of the number of parts in the buffer for a specific buffer size may be significantly different from that for another buffer size. This potential error can reduce the accuracy of the prediction model.
[0230]
3.4 Effective buffer capacity
[0231]
Increasing the capacity of the buffer affects one or more machines.
[0232]
3.4.1 Effect ratio
[0233]
18 to 21, the ratio of the time that a buffer affects another machine that is blocking or starving (the amount of decrease in blocking time or staging time) to the total time, as a percentage of the total time. Has been taken into account. This to total time ratio is always less than or equal to the total probability that the machine will be blocked or starved.
[0234]
On the other hand, there may be a need for a blocking time / starving time ratio (effect ratio), which is a ratio of the time at which one buffer affects another machine to the blocking / starving time.
[0235]
This ratio blocks the ratio of the time that one buffer affects the blocking or staging of another machine (decreased blocking time or reduced staging time) to the total time (total analysis time of the system). Or it can be calculated by dividing by the ratio of the time to be starved to the total time.
[0236]
For example, machine M3 is blocked at a 6.1 percent to total time ratio. Furthermore, increasing the capacity of the buffer AM3 reduces the blocking of the machine M3 at a 6.1% to total time ratio. Thus, the buffer AM3 reduces the blocking of the machine M3 with a 100% blocking time ratio. Although the buffer AM6 reduces the blocking of machine M3 to the total time ratio is only 2.8 percent, this corresponds to a 45.9 percent to blocking time ratio.
[0237]
In FIG. 34, the ratio at which machine M3 blocking decreases as the capacity in the buffer position increases is tabulated with respect to the total time ratio and the blocking time ratio. Machine M3 is blocked at a 6.1 percent to total time ratio, and, of course, if the buffer capacity in buffer AM3 increases, its blocking will be at a 6.1 percent to total time ratio, That is, it decreases with a 100 percent blocking time ratio.
[0238]
To determine the effect ratio that a buffer has on another machine that is blocked or starved, the total data shown in FIGS. 18-21 (the time that blocking or starving is reduced by adding empty space or parts) Ratio of total time) must be divided by the corresponding ratio of the time that the machine is starved or blocked to the total time.
[0239]
As represented by the equation in FIG. 35, the ratio PBEB-Ej, i (effect ratio) of the time for which the machine i is blocked when the blocking is reduced by the empty buffer j to the blocking time is the empty in the buffer j. Is calculated by dividing the ratio PBEBj, i of the time for which the blocking of machine i is reduced by the space to the total time by the ratio PM-Bi of the time for which machine i is blocked to the total time.
[0240]
The same is true for the ratio PBES-Ej, i (effect ratio) of the time that machine i is starved and its staging is reduced by empty buffer j to the staging time, and machine i is blocked. The ratio of the time that blocking is reduced by a full buffer j to the blocking time PBFB-Ej, i (effect ratio), and the time that the staring is reduced by a full buffer j as the machine i is starved This is true for the ratio PBFS-Ej, i (effect ratio) to the staging time.
[0241]
In FIG. 35, PBESj, i represents the ratio of the time that the staging is reduced by the empty buffer j as machine i is starved to the total time, and PM-Si is the total of the staging time. It represents the ratio to time.
[0242]
Further, in FIG. 35, PBFBj, i means the ratio of the time that machine i is blocked and its blocking is reduced by a full buffer j to the total time. Furthermore, PBFSj, i means the ratio of the time that machine i is starved and its staging is reduced by a full buffer j to the total time.
[0243]
3.4.2 Typical value of adjacent buffer capacity
[0244]
As described above, the average number of parts in the buffer is determined by the depletion rate PBFj where the buffer j is empty and the full rate PBEj where the buffer j is full. Here, they are the ratios of the buffer capacity change ΔBj and the effects of various buffers that are full and empty on the blocking and staging of multiple machines PBEB-Ej, i, PBES-Ej, i, PBFB- It is necessary to combine with Ej, i and PBFS-Ej, i.
[0245]
The typical value of the adjacent buffer capacity is
A representative additional empty space number ΔBEB-Aj, i behind machine i due to an increase in empty space in buffer j;
A representative number of additional parts ΔBES-Aj, i in front of machine i due to an increase in empty space in buffer j;
A representative number of additional empty spaces ΔBFB−Aj, i behind machine i due to the increase in parts in buffer j;
Due to the increase in the number of parts in buffer j, the number of representative additional parts ΔBFS−Aj, i in front of machine i and
Are distinguished. This is represented by the equations in FIG. 36 for all four cases.
[0246]
For example, when the capacity of the buffer j at the rear of the machine i is added, the representative additional empty space number ΔBEB-Aj, i of the buffer j is reduced when the empty space is added in the buffer j. It is calculated by the product of the ratio PBEB-Ej, i of the time to the blocking time, the depletion rate PBEj of the buffer j, and the capacity change amount ΔBj of the buffer j.
[0247]
Further, when the capacity of the buffer j in front of the machine i is added, the representative number of additional parts ΔBES-Aj, i of the buffer j is reduced by adding an empty space in the buffer j. It is calculated by the product of the ratio PBES-Ej, i of the time to the staging time, the depletion rate PBEj of the buffer j, and the capacity change amount ΔBj of the buffer j.
[0248]
For example, if the space of the buffer AM3 is increased by 4 and the buffer AM3 is usually 23% of its maximum capacity (see FIG. 27), an average of 3.08 additional empty spaces in the buffer. There is space and 0.92 additional parts.
[0249]
Further, the addition of capacity in buffer AM3 affects the blocking of machine M3 at a 100 percent time rate, and this additional capacity can be used at a 100 percent time rate. Thus, if there are 3.08 additional empty spaces in the buffer AM3, the blocking of the machine M3 will be affected at a time rate of 100 percent, i.e. the 3.08 spaces will reduce the blocking of the machine M2. 100% effect on.
[0250]
However, if the space of the buffer BM6 is increased by four, the buffer BM6 is always filled to 26.2% of the capacity, so on average, 2.96 spaces and 1.04 parts are added. The Furthermore, buffer BM6 affects the blocking of machine M3 only at a time rate of 77.05 percent, and the addition of 2.96 spaces is valid by a time rate of 77.05 percent. This can be expressed as adding only 2.30 spaces (2.96 x 77.05 percent) immediately after machine M3 to alleviate the blocking of machine M3.
[0251]
Evaluate a representative number of parts / spaces in a buffer adjacent to that machine by combining the percentage of time that one buffer affects the blocking of another machine with the increase in buffer size and the number of parts in the buffer. It is possible. This calculation needs to be done for the full effect that the modified buffer has on the various machines with respect to the various ways that one buffer can affect another machine, as shown below.
・ Mode of reducing blocking by adding free space in the buffer
A mode in which stubbing is reduced by adding free space in the buffer
-Aspect where blocking is reduced by the addition of available parts in the buffer
A mode in which stubbing is reduced by adding available parts in the buffer
[0252]
3.4.3 Combination of effective buffer capacity
[0253]
By converting the effect that all buffer increments (including the increase in free space and the number of parts) have on various machines to the typical buffer size of the (virtual) buffers adjacent to those machines The typical adjacent buffer capacity can be added for each machine, distinguishing whether it is a blocked machine or a starred machine.
[0254]
The additional amount ΔEi that adds available space behind machine i to reduce blocking is a typical addition behind machine i due to the increase in empty space in buffer j, as shown in the equation in FIG. The total value of the number of empty spaces ΔBEB−Aj, i and the number of representative additional empty spaces ΔBFB−Aj, i behind the machine i due to the increase in the number of parts in the buffer j.
[0255]
On the other hand, the additional amount ΔFi of adding available parts in front of machine i to reduce stubbing is that of machine i due to the increase in empty space in buffer j, as represented by the equation in FIG. This is the total value of the number of representative additional parts ΔBES−Aj, i in front and the number of representative additional parts ΔBFS−Aj, i in front of the machine i due to the increase in the number of parts in the buffer j.
[0256]
For example, in FIG. 4, increasing the space of buffer BM3 by 4 represents 0.90 empty space adjacent to machine M2, and increasing the space of buffer BM5 by 4 is adjacent to machine M2. Given 1.67 empty spaces, these empty spaces are combined to give 2.57 total empty space available to reduce machine M2 blocking.
[0257]
Similarly, increasing the space of the buffer BM3 by 4 and increasing the space of the buffer BM5 by 4 respectively means that 0.36 additional parts and 0.37 are available for the machine M6. This means that these available parts are combined, and the total number of parts that can be used by the machine M6 is 0.73. In this case, this increase in the number of available parts is due to two different effects.
[0258]
That is, first, due to the increase in available parts in the buffers BM3 and BM5, downstream staging including the machine M6 having a representative number of adjacent parts of (0.26 + 0.10) = 0.36 is reduced.
[0259]
In addition, more upstream empty space in the buffers BM3 and BM5 reduces the upstream blocking, which in turn reduces the stagnation of the machines M3, M4 and M5 in the parallel branch in the system shown in FIG. There is a possibility to make it.
[0260]
For example, if space for buffer BM3 is added, buffer BM3 is less likely to star machine M5 by blocking machine M2, so the available space in buffer BM3 can star machine M6. The nature is also lowered. Thus, the additional number of parts available to machine M6 is expressed as (0.27 + 0.10) = 0.37, and eventually the total number of additional parts available to mitigate machine M6 staging. The value is 0.73 (= 0.36 + 0.37).
[0261]
All of the additional adjacent empty spaces that are representative can be added for the machine, and moreover, all the additional adjacent parts that are representative can be added for the machine. Is possible. However, the adjacent additional empty space, which is representative, and the adjacent additional part, which is representative, have different effects and are not combined.
[0262]
After this calculation, each machine has a total value for all buffers j (j = 1 tom) of the number of typical additional spaces behind the machine and all buffers of the number of typical additional parts in front of the machine. and the total value for j (j = 1 tom).
[0263]
FIG. 38 shows a table of typical additional space numbers and representative additional component numbers for an example in which the buffer capacities of the buffers BM3 and BM5 are increased to five components. This table shows that there is 0.81 additional empty space behind machine M1, 2.57 additional empty space behind machine M2, and additional parts in front of machine M3. 2.49 and so on. In the next step, the number of additional spaces and the number of additional parts are converted into estimated values of reduced idle time.
[0264]
3.5 Reduction waiting time (reduction idle time)
[0265]
In the next step, reduced waiting time and reduced blocking time are determined for various machines when the buffer size is increased. This determination requires measurements of latency distribution for various machines.
[0266]
3.5.1 Unit action time TPP (Time Per Part)
[0267]
The basic problem associated with latency is how much time is saved when one piece of space is added for continued operation, one part for a starved machine and one for a blocked machine. That is. In this algorithm, the unit action time TPPi is assumed to be a time during which one part is in an action state, and thus includes a repair time and a tool change time. Unit action time TPPi for machine i is
TPPi = TBPi · PM-Ai
It is calculated using the following formula. That is, the unit action time TPPi is calculated by multiplying the average value of the unloading time interval TBPi of the machine i by the action time ratio P-MAi that is the ratio of the time during which the machine i is in the action state.
[0268]
The state of the machine i is classified into an operation state and a non-operation state. The operational state of the machine i includes an operating state of the machine i and other operational states, for example, a repair state and a tool change state. On the other hand, the non-operating state of the machine i includes a blocking state and a staging state, both of which are idle.
[0269]
The time during which the machine i is in the operating state is referred to as cycle time. Therefore, the unit action time TPP is expressed as the sum of the cycle time and the time when the machine i is in the action state except the operating state, and the unloading time interval TBP is a state where the unit action time TPP and the machine i are not acting. It is expressed as the sum of the time in
[0270]
In FIG. 39, the average carry-out time interval TBPi, the action time ratio PM-Ai, and the average unit action time TPPi are shown in a table for each machine. For example, when one available part is added to the starved machine M4, the machine M4 further operates for 103.4 s (seconds). Similarly, adding one available space to the blocked machine M1 will cause this machine M1 to operate for an additional 30s.
[0271]
3.5.2 Waiting time distribution
[0272]
The present algorithm is applied to all machine events, distinguishing starving time and blocking time from each other, so that the duration of the waiting time, which is starving time or blocking time, is determined. These data are expressed as a random distribution function, that is, a probability density function. The function is expressed as pdfM-Bi (t) and pdfM-Si (t) as probability density functions for the blocking and staging times of machine i. Their distribution is a function of time t.
[0273]
A standard distribution function (for example, an exponential function) may be used as the distribution function. However, it is usually easier and more accurate to define a special distribution function defined by the acquired data. That is, a probability density function pdf or a cumulative density function cdf can be defined by adapting a distribution function from a statistical textbook to the acquired data or simply using the acquired data to define this distribution. That's it. The approach of simply using the acquired data seems simple and accurate. In FIG. 40, a cumulative density function of a plurality of waiting times selected is shown in a graph.
[0274]
It is possible to estimate how much of the idle time will be reduced by adding available parts to reduce starving time or adding available space to reduce blocking time. is there. For example, when one free space is added to the machine M3 on average, the machine M3 can be operated for one cycle time, that is, 147s longer as shown in the above formula.
[0275]
If this information is combined with the distribution of the blocking time of the machine M3, it is possible to estimate the reduced blocking time. The equation shown in FIG. 41 is a mathematical calculation of the reduction ratio PΔTi of the unloading time interval TBPi for the machine i, and the distribution of the blocking time and the staging time, and the number of available spaces behind the machine i ΔEi. And the number of available parts ΔFi in front of the machine i, the ratio PM-Bi and PM-Si to the total time (total analysis time, designated analysis time, etc.) of blocking and starving time, and one part Is based on the unit action time TPPi, which is the time required to produce the. This equation can be expressed using probability density functions pdfM-Bi (t) and pdfM-Si (t).
[0276]
This equation consists of two terms, a reduction ratio term due to the reduced blocking time (first term) and a reduction ratio term due to the reduced staging time (second term).
[0277]
Each reduction ratio term gives the ratio of the time that a machine is blocked / starved to the ratio PM-Bi / PM-Si, the ratio of the average reduced blocking / starving time to the average total blocking / starving time. Calculated by multiplication.
[0278]
The average value of random variables in the system (for example, the staging time and blocking time of each machine i) is the product of the probability density function pdf and time t, between the lower end (usually 0 for latency) and the upper end. Can be determined by integrating over. If the upper end is less than infinity, the average value of the random variable is given for all times below this upper end. If the upper end is infinite, the average value of all random variables is given.
[0279]
The upper end can be determined as the product of the number of additional parts ΔFi for the staging time and the unit action time TPPi (also referred to as “cycle time” of the machine i), and the number of additional spaces ΔEi for the blocking time. , And can be determined as a product of the unit action time TPPi. The former product means the maximum amount that the working time of the machine i increases when a part is added for the machine i, and similarly, the latter product is that of the machine i when adding space for the machine i. It means the maximum amount that the action time increases.
[0280]
For example, an increase of 5 buffer BM3 and BM5 spaces will add 2.57 available space behind machine M2, which is an approximate reduction in unit action time TTP of 8.31 percent. It means that. The same buffer configuration shows that the additional number of parts available to machine M3 is 2.49, which reduces machine M3's staging by 5.29 percent.
[0281]
In the next step, the reduction rate ratio PΔTi of the unit action time TPPi is combined with the bottleneck probability and the initial system processing capacity in order to create a prediction model of the new processing capacity of the system. The total proportion of reduction waiting time for machine i is represented by PΔTi.
[0282]
3.6 Prediction model
[0283]
In order to determine the effective reduction rate ratio PΔTi of the unloading time interval TBP of the system and thus to determine the predicted value of the new unloading time interval TBP of this system, the reduction rate ratio PΔTi of the unloading time interval TBP for various machines. Is combined with the bottleneck probability.
[0284]
3.6.1 Effective reduction ratio of unloading time interval TBP
[0285]
In the above, the total reduction amount ratio PΔTi of the unloading time interval TBP is estimated for various machines. However, they are not independent of each other, but rather are part of the machine network. Thus, improving the performance of one machine may or may not affect the unloading time interval TBP of the system. The bottleneck probability PM-SBNi is used for each machine i to determine the ratio of the impact of improving the machine unload time interval TBP on the unload time interval TBP of the system. The effective total reduction ratio PΔT of the system is the total waiting time reduction ratio PΔTi for each machine i (that is, the reduction of the carry-out time interval TBP caused by the waiting time reduction amount, as shown by the equation in FIG. 42). The product of the quantity ratio) and the single bottleneck probability PM-SBNi is the sum of all machines (i = 1 ton).
[0286]
For example, the machine M1 may reduce the unloading time interval TBP to 24.18%. However, the machine M1 is by no means a bottleneck, and thus improving the machine M1 never improves the system take-out time interval TBP. Another machine, M3, is the bottleneck of this system and has a bottleneck probability of 32 percent.
[0287]
Therefore, 32% of the reduction waiting time of this machine M3 is an improvement in the unloading time interval TBP of this system. Since the estimated reduction latency is 5.3 percent, the effective reduction ratio is 32 percent of this value, which means that the overall system performance is 1.7 percent (= 32 percent x 1.7 percent). ) Improved.
[0288]
FIG. 43 shows a reduction waiting time, a single bottleneck probability of the machine, and a combination value of the machine unloading time interval TBP for the example in which the spaces of the buffers BM3 and BM5 are increased to five. The effective reduction amount is shown in the table. As a whole, the effective reduction amount ratio of the unloading time interval TBP can be totaled for all machines. In this example, the total carry-out time interval TBP may be reduced by 3.63 percent by improving the buffers BM3 and BM5 to have 5 space capacities. Theoretically, it is further possible to use the overall bottleneck probability rather than the single bottleneck probability. However, the prediction results were more accurate when using a single bottleneck probability.
[0289]
3.6.2 Improved Transport Time Interval TBP Prediction
[0290]
After calculating the effective reduction ratio PΔT of the carry-out time interval TBP of this system, it is very easy to predict a new value of the carry-out time interval TBP for the improved system. The formula for that is
TBP ′ = TBP · (1−PΔT)
In order to obtain the updated carry-out time interval TBP ′, the carry-out time interval TBP of the initial system is simply decreased by the effective reduction amount ratio PΔT of the carry-out time interval TBP calculated as described above.
[0291]
For example, increasing the capacity of the buffers BM3 and BM5 to 5 reduces the system export time interval TBP by 3.63 percent. Since the initial value of the system export time interval TBP was 55.68 s, the new value of the system export time interval TBP decreased by 2.02 s to 53.66 s. Thus, changing the space of buffers BM3 and BM5 from 1 to 5 reduces the total unload time interval TBP of this system by 3.63 percent from 55.68s to 53.66s.
[0292]
The theoretical background and specific method of this buffer prediction method have been described in detail above. Next, an example of a specific algorithm for implementing this buffer prediction method will be described with reference to a computer program.
[0293]
FIG. 44 conceptually shows a hardware configuration of the system analysis apparatus 10 used by the user to implement the buffer prediction method in a block diagram. The system analysis device 10 is an example of a buffer prediction device.
[0294]
As is well known, the system analysis apparatus 10 includes a computer 20 configured by connecting a processing unit (represented by “PU” in the figure) 12 and a storage 14 to each other via a bus 16. Yes.
[0295]
The computer 20 is connected to an input device 30 having a mouse and a keyboard as a pointing device, and an output device 40 that displays an image on a screen. The storage 14 is configured to include a recording medium such as a ROM, a RAM, a magnetic disk, and an optical disk. The user inputs necessary data to the computer 20 via the input device 30. In response to the input, the data processing result by the computer 20 is visualized and presented to the user via the output device 40.
[0296]
The storage 14 stores in advance a buffer prediction program executed by the PU 12 in order to implement this buffer prediction method. This buffer prediction program is designed to match the flow of the system on which it is executed (the relationship between the positions of the machines in the system and the positions of the buffers). In addition, the buffer prediction program is designed to meet the requirements for component flow in the system. Therefore, this buffer prediction program can reproduce the flow of parts in the system both in time and position as long as it is given system data that represents the change in the state of each component of the system over time. It is like that. Data used when the PU 12 executes the buffer prediction program is appropriately stored in the storage 14.
[0297]
FIG. 45 conceptually shows the contents of the buffer prediction program in a flowchart.
[0298]
In this buffer prediction program, first, a prediction model is created in step S100 (hereinafter simply expressed as “S100”, and the same applies to other steps). Details of this step are conceptually shown in a flowchart in FIG. 46 as a prediction model creation routine.
[0299]
In this prediction model creation routine, first, in S101, system data including the state of each machine / buffer at a given given time is acquired.
[0300]
This system data can be configured in the same format as in FIG. Specifically, the system data includes, for example, an element, that is, an identifier of a machine (for example, M1, M2, M3, etc.) and a state of the machine (for example, the above-described state 0, state 1, state 2, state 3, etc. ) And temporal information for specifying a period in which each machine indicates each state are related to each other. As will be described later, the time information is configured to include a start time and an end time of a period in which each machine indicates each state.
[0301]
Based on this system data, the operation time of each machine i is obtained by calculation from the start time and end time of each operation period. The system data can be, for example, data representing an actual operating state of the system in the past, or data representing a virtual operating state analyzed by simulation for the system. That is, this system data can be acquired, for example, by performing a single simulation or by measuring a production system.
[0302]
Next, in S102, some statistical values are analyzed. The statistical values include the average value of the unloading time interval TBPi of each machine i and the ratio (PM-Si and PM-Bi) to the total analysis time (designated analysis time) of the time when each machine i is in a specific state. And the average number of parts / spaces BFj and BEj in each buffer j.
[0303]
Subsequently, in S103, the bottleneck probability of each machine i is determined. The bottleneck probability can be determined, for example, by dividing the length of the working time of each machine i by the total analysis time, and one specific example thereof is the specification of the applicant's Japanese Patent Application No. 2002-81077. Detailed in the document.
[0304]
Thereafter, in S104, the distribution of blocking time and the distribution of staging time are determined for all machines. Thereby, probability density functions pdfM-Bi and pdfM-Si are defined for each machine i.
[0305]
This completes one execution of the prediction model creation routine.
[0306]
Thereafter, in S200 of FIG. 45, the cause of stubbing and blocking is analyzed. This detail is conceptually shown in the flowchart of FIG. 47 as a cause analysis routine.
[0307]
In this cause analysis routine, first, in S201, one of all the machines is selected as the first machine (starting element for tracking upstream or downstream). Next, in S202, any one of a plurality of events that can be taken by each machine is selected as an event to be noted for the first machine (hereinafter referred to as “the current machine event”).
[0308]
Subsequently, in S203, the weight is set to 1 and the current attention period is set to a period in which the current machine event continues.
[0309]
Thereafter, in S204, based on the system data, it is determined whether the state of the first machine is a blocking state, a staging state, or none of them.
[0310]
This time, assuming that the state of the first machine is a blocking state, the process proceeds to S600. In S600, the next downstream object of the first machine is determined.
[0311]
Details of S600 are conceptually shown in a flowchart in FIG. 48 as a downstream object determination routine.
[0312]
In the downstream object determination routine, first, in S601, all downstream objects adjacent to the current machine are ranked according to the following rule.
[0313]
Rank 1: Machine blocked
Rank 2: Full buffer
Rank 3: machine in action
Rank 4: Starved machine
Rank 5: Buffer not full
By this ranking, a path for generating the idle state of the first machine is determined for each downstream object in the system.
[0314]
Next, in S602, an object that is already part of the causal path leading to the causal element that caused the idle period of the first machine (this time is the blocking period) is analyzed. Excluded from. This is to prevent an infinite loop (deadlocking) from occurring when tracing each object downstream for cause analysis.
[0315]
Subsequently, in S603, among all the adjacent downstream objects described above, the one ranked in the highest order (there may be a plurality of objects) is selected. Thereafter, in S604, it is determined whether or not there is an object thus selected. If it is assumed that it does not exist this time, the determination is YES, immediately after one execution of this downstream object determination routine is completed, and then the process proceeds to S205 in FIG.
[0316]
On the other hand, this time, if it is assumed that the selected object exists, the determination in S604 in FIG. 48 is NO, and the process proceeds to S605.
[0317]
In S605, weights set in advance for the previous object (the case of the first machine, the object adjacent thereto, and the object following the object) may be set in advance. Is divided by the number of objects selected in S603 (may be plural). For example, for the bifurcated branch in the system, since the number of selected objects is 2, the weight of each object is set to 1/2. Thereafter, in S606, the first target object is selected from the highest-order target object (there may be a plurality of target objects).
[0318]
Subsequently, in S607, the period of interest is the next state change time of the object (there may be a plurality of objects) ranked highest from the start time of the state of the previous object, and the cause path It is set to the period up to the earlier of the end time of the previous object period above.
[0319]
Thereafter, in S608, it is determined whether or not the current order of the object is 4. If it is assumed that it is not 4 this time, the determination is no and the process moves to S800. In S800, the object is traced downstream for the set attention period. Details of S800 are conceptually shown in a flowchart in FIG. 50 as a downstream tracking routine.
[0320]
In this downstream tracking routine, first, in S801, it is determined whether or not the first machine is blocked. Since it is assumed that it is blocked this time, the determination is YES, and the process proceeds to S802.
[0321]
In S802, the weight set by execution of S605 or S705 described later (hereinafter simply referred to as “set weight”) and the attention period set by execution of S607 or S707 described later (hereinafter simply referred to as “set attention period”). )) (Which is simply referred to as “weighted time”) to the time when the first machine blocking is reduced by adding space in the buffer ahead of the current object. Is done. This time is related to the above-described mode 1 and is used to acquire the above-described ratio PBEB-Ej, i, that is, data (3) described later.
[0322]
After that, in S803, the current weighted time is a part of the path that caused the blocking of the first machine (the path between the first machine and the machine that caused the blocking). It was added to the time (which was located on the route). This time is used to acquire data (7) described later.
[0323]
Subsequently, in S804, it is determined whether or not the current target object is in an operating state. This time, if it is assumed that the current object is a machine and it is in an active state, the determination is YES, and in S805, the current object is the cause of blocking of the first machine. Determined.
[0324]
In S805, the current weighted time is the time that the current object causes blocking of the first machine (the current object continues to be the cause of causing blocking of the first machine). Time). This time is used for acquiring data (5) described later.
[0325]
Thus, one execution of the downstream tracking routine is completed, and the process proceeds to S609 in FIG.
[0326]
On the other hand, if it is assumed that the current object is not in the active state this time, the determination in S804 in FIG. 50 is NO, and the process proceeds to S806.
[0327]
In S806, the current weighted time is added to the time when the first machine blocking is reduced by adding a space in the buffer behind the current object.
[0328]
Thereafter, the process proceeds to S601 in FIG. 48, and the downstream object determination routine is executed again.
[0329]
Here, the execution contents of the downstream object determination routine of FIG. 48 and the downstream tracking routine of FIG. 50 will be described for the example shown in FIG.
[0330]
The downstream tracking routine of FIG. 50 is executed for machine M2, machine M3, and machine M4 because machine M1 is the first machine. The machine M1 is in the blocking state, and the downstream object determination routine of FIG. 48 first selects the machine M2 as the current object (the object following the machine M1).
[0331]
Thereafter, when the downstream tracking routine of FIG. 50 determines in S801 whether or not the first machine M1 is in the blocking state, the determination is YES, so in S802, the length of the period of the machine M2 (in S607) This weighted time, which is the product of the setting) and the weight (set in S605), is added to the time when the blocking state of the machine M1 decreases when a space is added in the buffer behind the machine M2. Further, in S803, the current weighted time is added to the time when the machine M2 was part of the path that caused the machine M1 to block.
[0332]
Since the machine M2 is not in the active state, the determination in S804 is NO, and then, in S806, if a space is added in the buffer behind the machine M2, the blocking state of the machine M1 decreases. It is added to the time to do.
[0333]
Thereafter, the process proceeds from the downstream tracking routine of FIG. 50 to the downstream object determination routine of FIG. 48, where the next downstream object is searched. The downstream object determination routine searches for the machine M3 as the current object, and therefore shifts again to the downstream tracking routine of FIG.
[0334]
In S801, the downstream tracking routine determines again whether or not the first machine M1 is in a blocking state. Since this determination is also YES this time, in S802, the length of the period of the machine M3 ( The current weighted time, which is the product of the setting (set in S607) and the weight (set in S605), is added to the time when the blocking state of the machine M1 decreases when a space is added in the buffer behind the machine M3. Further, in S803, the current weighted time is added to the time when the machine M3 was part of the path that caused the machine M1 to block.
[0335]
Since the machine M3 is not in the operating state, the determination in S804 is NO, and in S806, the current weighted time is the time that the blocking state of the machine M1 decreases when a space is added in the buffer behind the machine M3. Is added to
[0336]
Thereafter, the process proceeds from the downstream tracking routine of FIG. 50 to the downstream object determination routine of FIG. 48, where the next downstream object is searched. The downstream object determination routine searches for the machine M4 as the current object, and thus shifts again to the downstream tracking routine of FIG.
[0337]
This downstream tracking routine determines again whether or not the first machine, M1, is in a blocking state, and since this determination is also YES this time, in S802, the length of the period of the machine M4 (set in S607) ) And the weight (set in S605) is added to the time when the blocking state of the machine M1 decreases when a space is added in the buffer behind the machine M4. Further, in S803, the current weighted time is added to the time when the machine M4 was part of the path that caused the machine M1 to block.
[0338]
Since the machine M4 is in the operating state, the determination in S804 is YES, and in S805, the machine M4 is determined to be the original cause of the blocking of the first machine, the machine M1, and is weighted this time. Time is added to the time at which machine M4 causes blocking of machine M1.
[0339]
Thus, one execution of the downstream tracking routine is completed, and thereafter, the process proceeds to S609 in FIG. Since the machine M4 is the last object in the system, the determination in S609 is YES, and the process proceeds to S610.
[0340]
As described above, the case where the order of the target object is not 4 has been described, but when the target order is 4, the determination in S608 in FIG. 48 is YES, and in S900, the target object moves upstream in the set attention period. Followed.
[0341]
The details of S900 are conceptually shown in the flowchart of FIG. 51 as an upstream tracking routine. In this upstream tracking routine, first, in S901, it is determined whether or not the first machine is blocked. Since it is assumed that it is blocked this time, the determination is YES, and the flow proceeds to S902.
[0342]
In this S902, the current weighted time is added to the time when the blocking of the first machine is reduced if a part in the buffer behind the current object is added. This time is related to the above-described mode 3 and is used for acquiring the above-described ratio PBFB-Ej, i, that is, data (2) described later.
[0343]
Subsequently, in S903, the current weighted time is added to the time when the current object was part of the path that caused the blocking of the first machine. This time is used to acquire data (7) described later.
[0344]
After that, in S904, it is determined whether or not the current object is in the operating state. This time, if it is assumed that it is in the operating state, the determination is YES, and the process proceeds to S905. In S905, it is determined that the current object is the cause of blocking of the first machine.
[0345]
In S905, the current weighted time is added to the time at which the current object causes the first machine to block. This time is used for acquiring data (5) described later.
[0346]
Thus, one execution of this upstream tracking routine is completed, and thereafter, the process proceeds to S609 in FIG.
[0347]
On the other hand, this time, assuming that the current object is not in the active state, the determination in S904 in FIG. 51 is NO, and in S906, the current weighted time is ahead of the current object. Adding parts in the buffer adds to the time when the first machine blocking is reduced.
[0348]
Thereafter, the process proceeds to S701 in FIG. 49, and an upstream object determination routine described later is executed.
[0349]
In the downstream object determination routine of FIG. 48, when the execution of S800 or S900 is completed, it is then determined in S609 whether or not the iteration has been completed for all selected objects. If it is assumed that the process has not been completed yet, the determination is no, and the steps from S607 onward are executed for the next object.
[0350]
On the other hand, if it is assumed that the iteration has been completed for all the selected objects this time, the determination in S609 is YES, and the process proceeds to S610.
[0351]
In S610, it is determined whether or not the iteration until the entire period of the previous object state is covered is completed. If it is assumed that the process has not been completed yet, the determination is NO and the process returns to S601. However, if it is assumed that the process has been completed this time, the determination in S610 is YES, and one of the downstream object determination routines. Is completed, and then the process proceeds to S205 in FIG.
[0352]
Here, the execution contents of the downstream object routine of FIG. 48 will be specifically described taking the production system shown in FIG. 7 as an example.
[0353]
FIG. 68 is a time chart showing how the four machines M1 to M4 in the production system shown in FIG. 7 change between the blocking state and the working state along with the time t.
[0354]
The machine M1 is in a blocking state between time t0 and time t3. This buffer prediction program is executed to identify the cause of the blocking of the machine M1 and its path.
[0355]
Here, at the time when the computer 20 executes the downstream object determination routine of FIG. 48 in the buffer prediction program, the first machine or the previous object (that is, in any case, those Assume that the machine M1 is an object whose cause of the idle state is investigated), and a period between the time t0 and the time t3 is set as the attention period. During that period, the machine M1 is in a blocking state.
[0356]
In this case, the execution of S601 ranks all objects adjacent to the machine M1 on the downstream side thereof, that is, the machine M2 and the machine M3, and then the execution of S603 causes the machines M2 to be ranked. And it is determined that the machine M3 is given the same rank, and they are selected as the current object.
[0357]
When the machine M2 and the machine M3 are selected, the weight set for the machine M1 that is the previous object (that is, the weight of 1 set for the machine M1 in this example) is obtained by executing S605. Dividing by the number of objects ranked highest (ie, 2 in this example), the new weights for downstream objects M2, M3 will each be set as 1/2. .
[0358]
Thereafter, it is assumed that the execution of S606 has selected machine M2 as the first object with the top for further investigation. In this case, subsequently, by executing S607, the target period (the target period in which the current analysis is to be performed) becomes the highest level from time t0, which is the start time of the blocking period of the machine M1, which is the previous target object. Time of the next state change of the ranked objects (this time machines M2 and M3) (if there are multiple selected objects, it means the earliest time among them, in this example Among the machines M2 and M3 ranked at the top, the time t1 of the next state change of the machine M2 is applicable) and the end time t3 of the blocking period of the machine M1, which is the previous object. It is set as the period until. In this example, since the time t1 is earlier than the time t3, the current attention period is set as a period between the times t0 and t1, and the machine that is the current object is set for the set attention period. M2 will be tracked.
[0359]
Similar processing is performed for the machine M3 as the next object. The weight of the machine M3 is also set to 1/2 as described above. The attention period for the machine M3 is the time t1 of the next state change of the object ranked at the top (this time only the machine M3) from the time t0, and the blocking period of the machine M1, which is the previous object. Is set as the period up to the earlier of the end time t3. For this machine M3 as well as the machine M2, the current attention period is set as a period between times t0 and t1, and the machine M3 is tracked for the set attention period.
[0360]
The execution of S601 to S609 in FIG. 48 is repeated until the entire period of the specific state of the machine M1, which is the previous object, that is, the period from time t0 to time t3 is covered. Since the period from time t0 to time t1 has already been analyzed (covered), the period from time t1 to time t3 is now analyzed. When the machines M2 and M3, which are the downstream machines, are ranked for that period, the machine M2 is in the working state while the machine M3 is in the blocking state. It is higher than the ranking of M2. Therefore, for that period, there is only one top-level machine, and the weight of the machine M3 remains 1 because the number of objects ranked at the top is 1 this time. .
[0361]
The attention period (individual analysis period) for the machine M3 is the earlier of the time t2 of the next state change of the machine M3 and the end time t3 of the blocking period of the machine M1, which is the entire object, from the time t1. It is set as a period until a certain time t2. Therefore, the blocking of the machine M1 is tracked downstream for the period from time t1 to time t2. Since there is only one target ranked at the top this time, the determination in S609 is YES, and the process proceeds to S610.
[0362]
At present, both the period from time t0 to time t1 and the period from time t1 to time t2 are covered, and the remaining period is the period from time t2 to time t3. For this period, since both the machine M2 and the machine M3 are assumed to be in the working state, the same rank is given and the weight is set to ½.
[0363]
The execution of the loop constituted by S607, S608 and S800 is repeated for machine M2 and machine M3, and the execution of S800 tracks the blocking of machine M1 downstream. By executing S800, both the machine M2 and the machine M3 are determined to be the cause of the blocking of the machine M1 in the period from the time t2 to the time t3, and each of the machines M2 and M3 has a weight of 1. .
[0364]
If the analysis is performed for the entire period from time t0 to time t3, execution of the loop constituted by S601 to S609 ends.
[0365]
As described above, in the cause analysis routine of FIG. 47, it is assumed that the state of the first machine is the blocking state. However, if it is assumed that the state is the staging state, the next upstream object of the first machine is determined in S700. It is determined.
[0366]
Details of S700 are conceptually shown in a flowchart in FIG. 49 as an upstream object determination routine. This upstream object determination routine is basically the same as the downstream object determination routine of FIG.
[0367]
In this upstream object determination routine, first, in S701, all upstream objects adjacent to the first machine are ranked according to the following rules.
[0368]
Rank 1: Starved machine
Rank 2: empty buffer
Rank 3: machine in action
Rank 4: Machine blocked
Rank 5: Non-empty buffer
By this ranking, a path for generating the idle state of the first machine is determined for each upstream object in the system.
[0369]
Next, in S702, an object that is already part of the causal route leading to the causal element of the first machine idle period (this time, the staving period is assumed) is excluded. This is to avoid an infinite loop.
[0370]
Subsequently, in S703, the uppermost one of all the adjacent upstream objects described above is selected. Thereafter, in S704, it is determined whether or not there is an object selected in this way.
[0371]
If it is assumed that it does not exist this time, the determination is YES, and one execution of this upstream object determination routine is immediately terminated. On the other hand, this time, if it is assumed that the selected object exists, the determination is no and the process proceeds to S705.
[0372]
In S705, the weight is set to a value obtained by dividing the weight set in advance for the previous object by the number of objects selected in S703. Thereafter, in S706, the first object is selected from among the highest-order objects.
[0373]
Subsequently, in S707, the period of interest is the next state change time of the object (there may be a plurality of objects) ranked highest from the start time of the state of the previous object, and the cause path It is set to the period up to the earlier of the end time of the previous object period above.
[0374]
Thereafter, in S708, it is determined whether or not the order of the selected object is 4. If it is assumed that the current time is not 4, the determination is NO, and the object is traced upstream in S900 by executing the upstream tracking routine of FIG.
[0375]
In this upstream tracking routine, first, in S901, it is determined whether or not the first machine is blocked. Since it is assumed that the current state is the starving state this time, the determination is NO and the process proceeds to S807.
[0376]
In this step S807, the current weighted time is added to the time when the staging of the first machine is reduced by adding a space in the buffer ahead of the current object. This time is related to the above-described mode 2 and is used to acquire the above-described ratio PBES-Ej, i, that is, data (4) described later.
[0377]
Subsequently, in S808, the current weighted time is added to the time when the current object was part of the path that caused the first machine staging. This time is used for acquiring data (8) described later.
[0378]
Thereafter, in step S809, it is determined whether or not the current object is in an operating state. If it is assumed that it is in the working state this time, the determination is YES, and it is determined in S810 that the current object is the cause of the staging of the first machine.
[0379]
Further, in S810, the current weighted time is added to the time at which the current object causes the first machine to stave. This time is used for acquisition of data (6) described later.
[0380]
Thus, one execution of this downstream tracking routine is completed, and thereafter, the process proceeds to S609 in FIG.
[0381]
On the other hand, if it is assumed that the current object is not in the active state this time, the determination in S809 in FIG. 50 is NO, and the process proceeds to S811. In S811, the current weighted time is added to the time when the staging of the first machine is reduced by adding a space in the buffer behind the current object.
[0382]
Thereafter, the process proceeds to S601 in FIG. 48, and the downstream object determination routine is executed again.
[0383]
On the other hand, this time, assuming that the rank of the selected object is 4, the determination in S708 is YES, and in S800, the object is moved downstream by executing the downstream tracking routine of FIG. It is traced toward.
[0384]
In this downstream tracking routine, first, in S801, it is determined whether or not the first machine is blocked. Since it is assumed that the current state is the starving state this time, the determination is no and the process proceeds to S907.
[0385]
In this S907, the current weighted time is added to the time when the staging of the first machine is reduced if a part in the buffer behind the current object is added. This time is related to the above-described mode 4 and is used to acquire the above-described ratio PBES-Ej, i, that is, data (1) described later.
[0386]
Subsequently, in S908, the current weighted time is added to the time when the current object was part of the path that caused the first machine staging. This time is used for acquiring data (8) described later.
[0387]
Thereafter, in S909, it is determined whether or not the current target object is in an operating state. If it is assumed that the device is in the operating state this time, the determination is YES, and the flow proceeds to S910. In S910, it is determined that the current object is the cause of the staging of the first machine.
[0388]
Further, in S910, the current weighted time is added to the time at which the current object causes the first machine to stave. This time is used for acquisition of data (6) described later.
[0389]
Thus, one execution of this upstream tracking routine is completed, and thereafter, the process proceeds to S609 in FIG.
[0390]
On the other hand, if it is assumed that the current object is not in the active state this time, the determination in S909 in FIG. 51 is NO, and the process proceeds to S911. In S911, the current weighted time is added to the time when the staging of the first machine is reduced by adding a part in the buffer ahead of the current object.
[0390]
Thereafter, the process proceeds to S701 in FIG. 49, and an upstream object determination routine is executed.
[0392]
In either case, after that, in S709, it is determined whether or not the iteration has been completed for all the selected objects. If it is assumed that the process has not been completed this time, the determination is NO, and the steps after S707 are executed for the next object.
[0393]
On the other hand, if it is assumed that the iteration has been completed for all the selected objects this time, the determination in S709 is YES, and the process proceeds to S710.
[0394]
In S710, it is determined whether or not the iteration until the entire period of the previous object state is covered is completed. If it is assumed that the process has not been completed yet, the determination is NO and the process returns to S701. However, if it is assumed that the process has been completed this time, the determination in S710 is YES, and one of the upstream object determination routines. Is completed, and then the process proceeds to S205 in FIG.
[0395]
As described above, in the cause analysis routine of FIG. 47, the case where the state of the first machine is the stubbing state has been described, but if it is neither the blocking state nor the staging state, the process immediately proceeds to S205.
[0396]
In any case, in this S205, it is determined whether or not the iteration for all machine events has been completed. If it is assumed that the process has not been completed yet, the determination is no, and the next event is selected as the event for the first machine in S206, and then the process returns to S203. On the other hand, this time, if it is assumed that the iteration for all machine events is completed, the determination in S205 is YES.
[0397]
In this case, after that, in S207, it is determined whether or not the repetition for all the machines has been completed. If it is assumed that the process has not been completed yet, the determination is no, and after the next machine is selected as the first machine in S208, the process returns to S202. On the other hand, if it is assumed that the iterations for all machines have been completed this time, the determination in S207 is YES.
[0398]
After that, in S209, the buffer effect ratio that is the ratio of the effect due to the addition of the buffer / part, the cause component ratio that is the ratio that each machine constitutes the cause of the idle state of another machine, A path configuration ratio that is a ratio that configures a path that causes an idle state is calculated. This calculation is performed by appropriately dividing the length of a plurality of times determined so far (various times acquired by executing the downstream tracking routine of FIG. 50 or the upstream tracking routine of FIG. 51) by the total analysis time. Done.
[0399]
As the buffer effect ratio,
Data (1)
A ratio PBFS-Ej, i with respect to the total analysis time of the time when the stubbing of each machine i is reduced by adding the components in each buffer j,
Data (2)
If a part in each buffer j is added, a ratio PBFB-Ej, i of a time for blocking of each machine i to a total analysis time is reduced.
Data (3)
If the space in each buffer j is added, the ratio PBEB-Ej, i of the time for which the blocking of each machine i is reduced to the total analysis time,
Data (4)
If the space in each buffer j is added, the ratio of the time for which the staring of each machine i is reduced to the total analysis time PBES-Ej, i
Is included.
[0400]
On the other hand, as the cause component ratio,
Data (5)
The ratio of the time that one machine was responsible for blocking another machine to the total analysis time,
Data (6)
The ratio of the time that one machine was responsible for starving another machine to the total analysis time
Is included.
[0401]
In addition, as the route composition ratio,
Data (7)
The ratio of the time that one machine was in the path of another machine's block to the total analysis time,
Data (8)
The ratio of the time that one machine was on the stave path of another machine to the total analysis time
Is included.
[0402]
However, the ratios of the data (1) to (4) are calculated for all combinations of the position of the buffer j and the machines i before and after each machine i, while the ratios of the data (5) to (8) are calculated. ) Ratio is calculated for all combinations of machines i.
[0403]
Thus, one execution of the cause analysis routine is completed, and thereafter, the process proceeds to S300 in FIG. In S300, additional buffer capacities for buffers before and after the machine are specified. Subsequently, in S400, the effect of the additional buffer capacity is estimated. Details of S400 are conceptually shown in a flowchart in FIG. 52 as a buffer effect estimation routine.
[0404]
In this buffer effect estimation routine, first, in S1000, the average number of parts in the buffer in front and rear of each machine is estimated. Details of S1000 are conceptually shown in a flowchart in FIG. 53 as an average component number estimation routine.
[0405]
In this average number-of-parts estimation routine, first, in S1001, the ratio of the time during which another machine was part of the path of the current machine being starved to the total analysis time (the above data (8)) is set. , The bottleneck probability of this machine is multiplied, thereby calculating the weighted upstream bottleneck probability.
[0406]
Next, in S1002, the ratio of the time when another machine is part of the path of the current machine in the block to the total analysis time (the above-mentioned data (7)) shows the bottleneck probability of the current machine. Multiply, thereby calculating the weighted downstream bottleneck probability.
[0407]
Subsequently, in S1003, it is determined whether or not the other machine has covered the entire machine, that is, whether or not the other machine has been changed within the range of the entire machine. If it is assumed that the entire machine is not covered this time, the determination is NO, and the steps after S1001 are executed again for another updated machine. On the other hand, if it is assumed that all machines are covered this time, the determination in S1003 is YES, and the process proceeds to S1004.
[0408]
In S1004, several weighted upstream bottleneck probabilities for several machines are summed to calculate a total weighted upstream bottleneck probability. Further, the total weighted downstream bottleneck probabilities for several machines are summed to calculate the total weighted downstream bottleneck probability.
[0409]
Subsequently, in S1005, the average number of parts BFj in the buffer j in front of the current machine is calculated. In particular,
BFj = buffer capacity × total weighted upstream bottleneck probability / (total weighted upstream bottleneck probability + total weighted downstream bottleneck probability + current machine bottleneck probability)
It is calculated using the following formula.
[0410]
Thereafter, in S1006, the average number of parts BFj in the buffer j behind the current machine is calculated. In particular,
BFj = buffer capacity x (total weighted upstream bottleneck probability + current machine bottleneck probability) / (total weighted upstream bottleneck probability + total weighted downstream bottleneck probability + current machine bottleneck probability)
It is calculated using the following formula.
[0411]
Subsequently, in S1007, it is determined whether or not the iteration has been completed for all the machines. If it is assumed that the process has not been completed this time, the determination is no, and the steps after S1001 are executed again for the updated machine. On the other hand, if it is assumed that the iteration has been completed for all machines this time, the determination in S1007 is YES, and one execution of this average component number estimation routine is completed, and then S1100 in FIG. Migrate to
[0412]
In addition, in S102 of FIG. 46, the average number of parts in the buffer is calculated. This point is common to this average part number estimation routine. However, in S102, the average number of parts in the buffer that actually exists in the simulation model is measured. On the other hand, in this average part number estimation routine, the average number of parts in the virtual buffer is measured. It is estimated that the actual existence of a buffer with which the average number of parts is associated is not required.
[0413]
In S1100, the number of additional parts and spaces available to the machine is estimated. Details of S1100 are conceptually shown in a flowchart of FIG. 54 as an additional component / space number estimation routine.
[0414]
In this additional part / space number estimation routine, first, in S1101, if a part is added to the current buffer j, the ratio of time PBFSj, i (the above-mentioned data (1)) ) Is multiplied by the average number of parts BFj in the buffer j and added to the number of additional parts ΔFi for the current machine i.
[0415]
Next, in S1102, if the space is added to the current buffer j, the time ratio PBESj, i (the above-mentioned data (4)) in which the staring of the current machine i is reduced is added to the average number of spaces BEj in the buffer j. Is added to the number of additional parts ΔFi for the current machine i.
[0416]
Subsequently, in S1103, if the part is added to the current buffer j, the average ratio BFj in the buffer j is equal to the time ratio PBFBj, i (the above-mentioned data (2)) in which blocking of the current machine i is reduced. Multiply and add to the additional space number ΔEi for machine i this time.
[0417]
Thereafter, in S1104, the time ratio PBEBj, i (the above-mentioned data (3)) in which the blocking of the current machine i is reduced if a space is added to the current buffer j is multiplied by the average number of spaces BEj in the buffer j. It is added to the number of additional spaces ΔEi for the current machine i.
[0418]
Subsequently, in S1105, it is determined whether or not the iteration has been completed for all the buffers, that is, whether or not updating of the buffer j is unnecessary. If it is assumed that the iteration has not been completed for all the buffers this time, the determination is no, the buffer j is updated, and the process returns to S1101. On the other hand, if it is assumed that the iteration has been completed for all the buffers, the determination in S1105 is YES, and the process proceeds to S1106.
[0419]
In S1106, it is determined whether or not the iteration has been completed for all the machines, that is, whether or not the update of the machine i is unnecessary. If it is assumed that the iteration has not been completed for all the machines this time, the determination is NO, and after the machine i is updated, the process returns to S1101. On the other hand, if it is assumed that the iteration has been completed for all the machines, the determination in S1106 is YES.
[0420]
Thus, one execution of this additional part / space number estimation routine is completed, and thereafter, the process proceeds to S1200 in FIG.
[0421]
In S1200, the reduced idle time ratio of the machine is estimated. Details of S1200 are conceptually shown in a flowchart of FIG. 55 as a reduced idle time ratio estimation routine.
[0422]
In this reduced idle time ratio estimation routine, first, in S1201, when a part is added for the current machine i, the maximum increase amount, which is the maximum amount that the working time of the current machine i increases, This is determined by multiplying the number of additional parts ΔFi for this by the unloading time interval TPPi of the current machine i.
[0423]
Next, in S1202, the average stubbing time is determined by integrating the current staging time distribution function of the machine i with respect to all the domains thereof.
[0424]
Subsequently, in S1203, by integrating the staging time distribution function of the current machine i with respect to the range of 0 and the maximum increase amount ΔFi · TPPPi of the working time due to the component addition, the reduced stubbing time is calculated. It is determined.
[0425]
Thereafter, in S1204, the ratio PM-Si of the time when the current machine i is starved to the total analysis time is multiplied by the ratio of the reduced staging time and the average staging time, thereby obtaining the current machine i's ratio. Therefore, when a part is added, a reduced idle time ratio, which is a ratio of time during which idle time is reduced, is determined.
[0426]
Subsequently, in S1205, when a space is added for the current machine i, the maximum increase amount, which is the maximum amount that the working time of the current machine i increases, is added to the additional space number ΔEi for the current machine i. It is determined by multiplying the unloading time interval TPPi of the current machine i.
[0427]
Thereafter, in S1206, the average blocking time is determined by integrating the blocking time distribution function of the current machine i with respect to all the domains thereof.
[0428]
Subsequently, in S1207, the blocking time distribution function of the current machine i is integrated with respect to the range between 0 and the maximum increase amount ΔEi · TPPi of the action time caused by the space addition, thereby reducing the blocking time. It is determined.
[0429]
Thereafter, in S1208, the ratio of the reduced blocking time to the average blocking time is multiplied by the ratio PM-Bi of the time when the current machine i is blocked to the total analysis time, thereby obtaining the space for the current machine i. Is added, a reduced idle time ratio, which is a ratio of time during which idle time is reduced, is determined.
[0430]
Subsequently, in S1209, the reduced idle time ratio PΔTi for the current machine i as a whole is determined by summing the reduced idle time ratio by adding a space and the reduced idle time ratio by adding a part.
[0431]
Thereafter, in S1210, it is determined whether or not the iteration has been completed for all the machines. If it is assumed that the process has not been completed this time, the determination is NO and the process returns to S1201 for the next machine i. However, if it is assumed that the process has been completed this time, the determination is YES.
[0432]
Thus, one execution of this reduced idle time ratio estimation routine is completed, and thereafter, the process proceeds to S401 in FIG.
[0433]
In S401, for the current machine i, the effective idle time ratio PΔTi is determined by multiplying the reduced idle time ratio PΔTi by the bottleneck probability PM-SBNi of the current machine i.
[0434]
Thereafter, in S402, it is determined whether or not the iteration has been completed for all the machines. If it is assumed that the process has not been completed this time, the determination is no, and the process proceeds to S401 for the next machine i. On the other hand, if it is assumed that the iteration has been completed for all machines this time, the determination in S402 is YES, and the process proceeds to S403.
[0435]
In S403, the determined effective reduction idle time ratio PΔTi is totaled for all the machines, thereby determining the improvement amount PΔT of the entire system.
[0436]
Subsequently, in S404, the improved system carry-out time interval TBP 'is estimated by reducing its previous carry-out time interval TBP by the improvement amount PΔT.
[0437]
Thereafter, in S405, the in-process inventory quantity of the improved system is the sum of the average value of the average number of parts in all the buffers and the average number of parts in the machine corresponding to the ratio of the time for which the machine accommodates the parts. Presumed.
[0438]
Subsequently, in S406, the improved system total processing time (makespan) is estimated as the product of the improved system in-process inventory and the improved system take-out time interval TBP '.
[0439]
Thus, one execution of this buffer effect estimation routine is completed, and thereafter, the process proceeds to S500 in FIG.
[0440]
In S500, it is determined whether or not the system is optimized based on the estimated total processing time and the like. If it has not been optimized, the determination is no, the process returns to S300, and a new additional buffer capacity is designated. On the other hand, this time, if it is assumed that the system has been optimized, the determination in S500 is YES, and one execution of this buffer prediction program ends.
[0441]
As is clear from the above description, in this embodiment, S201, S204, S600, and S700 constitute an example of the “first tracking step” in the item (1), and S804, S805, S809, S810, S904, S905, S909, and S910 constitute an example of the “cause identification step” in the same term, S600 and S700 constitute an example of the “second tracking step” in the term, and S601, S603, S606, S701, S703 and S706 constitute an example of the “route specifying step” in the same section.
[0442]
Furthermore, in the present embodiment, S204, S608, and S708 constitute an example of the “orientation setting step” in the item (3).
[0443]
Furthermore, in this embodiment, S601 and S701 constitute an example of the “ranking step” in the above section (4), and S603, S606, S608, S703, S706, and S708 are “identifying each cause route”. It constitutes an example of “step to do”.
[0444]
Further, in the present embodiment, S800 and S900 constitute an example of the “statistical value calculation step” in the above item (5).
[0445]
Further, in the present embodiment, S1000 constitutes an example of the “first estimation step” in the above item (6), S1100 constitutes an example of the “second estimation step” in the same term, and S1200 corresponds to the same term. An example of the “third estimation step” is configured, and S401 to S406 configure an example of the “fourth estimation step” in the same section.
[0446]
Furthermore, in the present embodiment, S1100 constitutes an example of the “second estimation step” in the item (7).
[0447]
Furthermore, in the present embodiment, S1000 constitutes an example of the “first estimation step” in the item (8).
[0448]
Furthermore, in the present embodiment, the buffer prediction program constitutes an example of the “program” according to the item (9), and the storage 14 that stores the buffer prediction program in advance includes the “recording medium” according to the item (10). "Is an example.
[0449]
Next, the buffer prediction method described above is verified by comparing the predicted value with the measured value.
[0450]
In this verification, the predicted improvement in system throughput is simply compared to the actual improvement for various prediction models. In general, the predicted value is not always perfect, but usually has practical accuracy.
[0451]
1 Linear production system with 8 machines
[0452]
This is a simple example of a linear production system with 8 machines numbered from 1 to 8. Buffer positions exist between the machines, and numbers A to G are assigned as shown in FIG. The main bottleneck is machine M4 with a total bottleneck probability of 95 percent. Two buffers adjacent to the bottleneck, buffer C and buffer D, were analyzed. Since the performance of this system is mainly determined by the machine M4, the prediction of this model is very easy. Machine M4 has a much larger bottleneck probability and the two analyzed buffers C and D are adjacent to the bottleneck. This greatly simplifies the prediction model and greatly increases its prediction accuracy.
[0453]
1.1 Buffer C
[0454]
FIG. 57 shows a comparison between the predicted value of the system processing capacity (unloading time interval TBP) obtained by using this buffer prediction method and the measured data on the buffer size of the buffer C in which the space changes from 0 to 10. ing. Measurement data is shown with a 95 percent confidence interval for throughput. Overall, the predicted value follows the measured data very approximately, and almost all predicted values are within the 95 percent confidence interval of the measured data. The square root of the mean total square error was only 0.25 s.
[0455]
1.2 Buffer D
[0456]
FIG. 58 shows a comparison between the predicted value of the processing capacity (export time interval TBP) obtained by using this buffer prediction method and the measured data on the buffer size of the buffer D in which the space changes from 0 to 5. Yes. Due to the short-term simulation and the small buffer effect, the confidence interval is slightly larger than the overall improvement. Nevertheless, the predicted value is almost always within the 95 percent confidence interval, and the predicted value is very accurate.
[0457]
2 Branch production system with 7 machines
[0458]
An example of this is shown in FIG. Buffers AM2, BM3, AM3 and AM4 were investigated in detail. Buffers BM5 and AM5 were briefly investigated.
[0459]
2.1 Buffer AM2
[0460]
FIG. 59 shows a comparison between the predicted value of the processing capacity (export time interval TBP) obtained by using this buffer prediction method and the measured data on the buffer size of the buffer AM2 in which the space changes from 0 to 10. Yes. Measurement data is shown with a 95 percent confidence interval for throughput. Overall, the predicted value follows the measured data very well in the early stages, and has a small error for values with a large buffer size. The square root of the mean total square error was 0.40 s.
[0461]
2.2 Buffer BM3
[0462]
FIG. 60 shows a comparison between the predicted value of the processing capacity (export time interval TBP) obtained by using this buffer prediction method and the measured data on the buffer size of the buffer BM3 in which the space changes from 0 to 10. Yes. Measurement data is shown with a 95 percent confidence interval for throughput. The predicted value follows the measured data very well, and almost all predicted values are within the 95 percent confidence interval of the measured data. The square root of the mean total square error was only 0.22 s.
[0463]
2.3 Buffer AM3
[0464]
FIG. 61 shows a comparison between the predicted value of the processing capability (export time interval TBP) obtained by using this buffer prediction method and the measurement data on the buffer size of the buffer AM3 in which the space changes from 0 to 10. Yes. Measurement data is shown with a 95 percent confidence interval for throughput. The predicted value follows the measured data very well for the first few values of the buffer size. The square root of the mean total square error was 0.78 s.
[0465]
2.4 Buffer AM4
[0466]
FIG. 62 shows a comparison between the predicted value of the processing capability (export time interval TBP) obtained by using this buffer prediction method and the measured data on the buffer size of the buffer AM4 in which the space changes from 0 to 10. Yes. Measurement data is shown with a 95 percent confidence interval for throughput. The square root of the mean total square error was 0.96 s.
[0467]
2.5 Buffer BM5
[0468]
FIG. 63 shows a comparison between the predicted value of the processing capability (export time interval TBP) obtained using this buffer prediction method and the measured data on the buffer size of the buffer BM5 in which the space changes from 0 to 10. Yes. However, this buffer BM5 has been investigated only briefly, and verification was performed only for the case of 5 in addition to the initial value of the buffer size. The predicted value is within the 95 percent confidence interval of the measured data and is therefore valid. Due to the small number of data, there is no square root of the mean total square error.
[0469]
2.6 Buffer AM5
[0470]
FIG. 64 shows a comparison between the predicted value of the processing capability (export time interval TBP) obtained by using this buffer prediction method and the measured data on the buffer size of the buffer AM5 in which the space changes from 0 to 10. Yes. Again, this buffer AM5 has only been investigated easily, and verification has been performed only for the case of 5 in addition to the initial value of the buffer size. The predicted value is within the 95 percent confidence interval of the measured data and is therefore valid. Due to the small number of data, there is no square root of the mean total square error.
[0471]
2.7 Summary
[0472]
This predictive model appears to give accurate results for most cases. The buffers AM2, BM3, BM5 and AM5 are very accurate. The initial trend in buffer AM3 was well predicted.
[0473]
In addition to these experiments, a simulation was added once for a new combination of buffers. The buffers AM2, BM3, AM3, AM4, BM5 and AM5 were each set to have a slightly large capacity, such as 20 spaces. The simulation determined that a carry-out time interval TBP of 48.89 s had a confidence interval of 0.19 s on each side. The predicted value is 50.69 s, which is never wrong. There is a difference of about 0.8 s between them, but this error is small relative to the initial value of 55.7 s and the trend is very well predicted. The total square root of the mean total square error was 0.82 s.
[0474]
3 A system where two machines branch
[0475]
As shown in FIG. 65, in this example, two machines and three buffers are provided. Type A parts and Type B parts arrive at the buffer BM0. There are two type B parts for each type A part that arrives, and these parts are supplied indefinitely. All type A parts go to machine M1 via buffer BM1, and all type B parts go to machine M2 via buffer BM2. Machine M1 is the primary bottleneck with a 90% total bottleneck probability and machine M2 is a secondary bottleneck with a 30% total bottleneck probability.
[0476]
Two important buffers are buffers BM1 and BM2. These buffers can star subsequent machines and can also block machines in parallel paths. For example, a full buffer BM1 may cause buffer BM0 to block and thus may cause BM2 to star. Of course, the empty buffer BM1 staves the machine M1.
[0477]
3.1 Buffer BM1
[0478]
FIG. 66 shows a comparison between the predicted value of the processing capability (export time interval TBP) obtained using this buffer prediction method and the measured data on the buffer size of the buffer BM1 in which the space changes from 0 to 10. Yes. Measurement data is shown with a 95 percent confidence interval for throughput. Overall, the predicted value follows the measured data very well in the early stages, and has a small error for values with a large buffer size. The square root of the mean total square error was 0.44 s.
[0479]
3.2 Buffer BM2
[0480]
FIG. 67 shows a comparison between the predicted value of the processing capacity (export time interval TBP) obtained by using this buffer prediction method and the measured data on the buffer size of the buffer BM2 in which the space changes from 0 to 10. Yes. Measurement data is shown with a 95 percent confidence interval for throughput. The square root of the mean total square error was 0.56 s.
[0481]
3.3 Summary
[0482]
This predicted value is not always satisfactory for large changes in buffer size. The total square root of the mean square error for those data was 0.51 s. In addition to the above data values, another simulation was performed with two very large buffers having 50 spaces. The measured value of the export time interval TBP was 36.50 s (plus or minus 0.14 s for the 95 percent confidence interval), which is practically 36.81 s for a large range of predicted values. It is close to the predicted value.
[0483]
4 Summary
[0484]
Overall, this buffer prediction method works well in practice for small changes in buffer size and works very well for linear systems.
[0485]
By the way, when it is necessary to improve the size of the buffer between each component in the system, the conventional trial-and-error exploratory method using simulation simulates at least whenever the capacity (size) of the buffer is changed. It was necessary to change the model settings for and to perform simulation again. Therefore, a result cannot be obtained unless two waiting times, that is, the time required to change the model and the execution time of the simulation have elapsed.
[0486]
In addition, in the case of many exploratory methods, the optimal solution is only derived as a search result, but the optimal solution has an effect on how sensitive each buffer is to the evaluation function. It does not indicate what. For this reason, it has often been difficult for designers to understand the solution, that is, the appropriateness of buffer improvement measures.
[0487]
On the other hand, according to the present embodiment, it is possible to instantaneously predict the effect brought about by merely indicating how to improve which buffer. Therefore, the designer can repeat trial and error in design without stopping his thought process.
[0488]
As mentioned above, although one embodiment of the present invention was described in detail based on a drawing, this is an illustration, and it is based on the knowledge of those skilled in the art including the aspect described in the section of [Means for Solving the Problems]. The present invention can be implemented in other forms based on various modifications and improvements.
[Brief description of the drawings]
FIG. 1 is a flowchart conceptually showing a buffer prediction method according to an embodiment of the present invention.
FIG. 2 is a time chart for explaining an example in which the state of each machine in the production system in which the buffer prediction method of FIG. 1 is implemented changes with time;
FIG. 3 is a diagram for explaining an example of system data that needs to be used to implement the buffer prediction method of FIG. 1 in a table format;
4 is a flowchart conceptually showing an example of a production system in which the buffer prediction method of FIG. 1 is implemented.
FIG. 5 is a flowchart schematically showing a buffer prediction model used by the buffer prediction method of FIG. 1;
FIG. 6 shows a system with three machines and two buffers on a straight line.
FIG. 7 is a diagram showing a system including four machines on a line having a branch portion.
FIG. 8 is a diagram illustrating an example of a system in which reverse blocking occurs.
FIG. 9 is a diagram illustrating an example of a system in which reverse stubbing occurs.
10 is a diagram for explaining the picture display in FIGS. 11 and 12. FIG.
FIG. 11 is a diagram for explaining five stubbing situations that the machine Mi can take;
FIG. 12 is a diagram for explaining five blockin situations that the machine Mi can take;
13 is a graph showing the probability of stubbing occurring in each machine in the production system of FIG.
14 is a graph showing the probability that blocking will occur in each machine in the production system of FIG. 4; FIG.
15 is a diagram for explaining the cause of blocking and stubbing in the machine M3 in the production system of FIG. 4;
16 is a diagram for explaining the cause of blocking and staging in the machine M5 in the production system of FIG. 4;
FIG. 17 is a diagram showing, in tabular form, modes 1 to 4 regarding the effect that a buffer has on a machine in the system.
FIG. 18 is a diagram showing, in a tabular form, a time ratio at which a certain machine is blocked in the mode 1;
FIG. 19 is a diagram showing, in tabular form, a time ratio at which a certain machine is stared in the mode 2;
FIG. 20 is a diagram showing, in a tabular form, a time ratio at which a certain machine is blocked in the mode 3;
FIG. 21 is a diagram showing, in tabular form, a time ratio at which a certain machine is stared in the mode 4;
FIG. 22 shows several formulas related to the effective buffer size.
FIG. 23 is a histogram showing a probability that a part exists in a buffer AM3 having a size of 1 in the production system of FIG. 4;
24 is a histogram showing the probability that a part exists in a buffer AM3 having a size of 10 in the production system of FIG.
25 is a table showing the bottleneck probability of each machine in the initial setting of the production system of FIG. 4 in a table format.
FIG. 26 is a diagram for explaining distribution of bottleneck probabilities on the downstream side of the buffer AM1 in the production system of FIG. 4;
27 is a diagram showing, in a tabular form, a predicted value and a measured value of the fullness rate of each buffer in the production system of FIG. 4 in contrast.
FIG. 28 is a graph showing the predicted value and the measured value of the fullness rate of each buffer before capacity change in the production system of FIG. 4 in comparison.
29 is a diagram showing, in a tabular form, the bottleneck probability of each machine for the production system of FIG. 4 before and after the change of the buffer capacity.
30 is a diagram showing, in a tabular form, the predicted value and the measured value of the fullness rate of each buffer in the production system of FIG. 4 before and after the change of the buffer capacity.
FIG. 31 is a graph showing the predicted value and the measured value of the fullness rate of each buffer after capacity change in comparison with the production system of FIG. 4;
32 is a histogram showing a probability that a part exists in a buffer AM4 having a size of 5 in the production system of FIG. 4;
33 is a graph showing the average number of parts and the average part number ratio of the buffer AM4 in the production system of FIG. 4;
FIG. 34 is a diagram showing, in a tabular form, a total time ratio and a blocking time ratio that are affected by the blocking of the machine M3 when a space is added to each buffer in the production system of FIG.
FIG. 35 illustrates several equations related to the effect ratio by adding space or parts in the buffer.
FIG. 36 is a diagram illustrating several expressions related to the representative value of the adjacent buffer capacity.
FIG. 37 is a diagram illustrating several expressions related to a composite value of effective buffer capacities.
FIG. 38 is a diagram showing, in a tabular form, the number of representative additional spaces and the number of representative additional parts of each machine when the capacity of the buffers BM3 and BM5 is increased to 5 in the production system of FIG.
FIG. 39 is a diagram showing, in a tabular form, carry-out part intervals, action time ratios, and unit action times of each machine in the production system of FIG.
FIG. 40 is a graph showing the characteristics of the cumulative distribution function for the waiting time of several selected machines for the production system of FIG. 4;
FIG. 41 is a diagram illustrating an expression related to a reduction waiting time for each machine.
FIG. 42 is a diagram illustrating an expression related to a reduction waiting time for all machines.
43 is a table showing the reduction waiting time ratio, the single bottleneck probability, and the effective reduction amount ratio of each machine in the production system of FIG. 4 in a table format.
FIG. 44 is a block diagram showing a hardware configuration of a system analysis apparatus suitable for carrying out the buffer prediction method.
45 is a flowchart conceptually showing the contents of a buffer prediction program in FIG. 44. FIG.
FIG. 46 is a flowchart conceptually showing S100 in FIG. 45 as a prediction model creating routine.
FIG. 47 is a flowchart conceptually showing S200 in FIG. 45 as a cause analysis routine.
FIG. 48 is a flowchart conceptually showing S600 in FIG. 47 as a downstream object determination routine.
FIG. 49 is a flowchart conceptually showing S700 in FIG. 47 as an upstream object determination routine.
FIG. 50 is a flowchart conceptually showing S800 in FIGS. 48 and 49 as a downstream tracking routine.
FIG. 51 is a flowchart conceptually showing S900 in FIGS. 48 and 49 as an upstream tracking routine.
FIG. 52 is a flowchart conceptually showing S400 in FIG. 45 as a buffer effect estimation routine.
FIG. 53 is a flowchart conceptually showing S1000 in FIG. 52 as an average component number estimation routine.
FIG. 54 is a flowchart conceptually showing S1100 in FIG. 52 as an additional component / space number estimation routine.
FIG. 55 is a flowchart conceptually showing S1200 in FIG. 52 as a reduced idle time ratio estimation routine.
FIG. 56 is a diagram showing a linear production system having eight machines.
57 is a graph showing the predicted value and measured value of the carry-out time interval TBP of the buffer C in FIG. 56 in comparison.
58 is a graph showing the predicted value and measured value of the export time interval TBP of the buffer D in FIG. 56 in comparison.
59 is a graph showing a comparison between a predicted value and a measured value of the carry-out time interval TBP of the buffer AM2 in FIG.
60 is a graph showing a comparison between a predicted value and a measured value of the unloading time interval TBP of the buffer BM3 in FIG.
61 is a graph showing a comparison between a predicted value and a measured value of the carry-out time interval TBP of the buffer AM3 in FIG.
62 is a graph showing a comparison between a predicted value and a measured value of the carry-out time interval TBP of the buffer AM4 in FIG.
63 is a graph showing a comparison between a predicted value and a measured value of the unloading time interval TBP of the buffer BM5 in FIG.
64 is a graph showing a comparison between a predicted value and a measured value of the carry-out time interval TBP of the buffer AM5 in FIG.
FIG. 65 is a diagram showing a production system in which two machines are branched.
66 is a graph showing a comparison between a predicted value and a measured value of the carry-out time interval TBP of the buffer BM1 in FIG. 65. FIG.
67 is a graph showing a comparison between a predicted value and a measured value of the carry-out time interval TBP of the buffer BM2 in FIG. 65. FIG.
68 is a time chart showing an example of a change over time in the state of each machine in the production system of FIG. 7. FIG.
FIG. 69 is a diagram for describing definitions of some symbols used in the description of the embodiment in a tabular format.
FIG. 70 is a diagram for describing another symbol definition used in the description of the embodiment in a table format;
[Explanation of symbols]
10 System analyzer
12 Processing Unit
14 Storage
20 computers

Claims (8)

In order to generate a flow from the upstream side to the downstream side of the processed element, a system in which a plurality of components to be processed with respect to the processed element are arranged according to a predetermined flow. A computer having a processor and a storage includes a cause element that is another component that causes an idle state in a certain component and a cause path from the cause element to the idle component that is the idle state component. A cause / path estimation method for estimation,
A step in which the processor reads from the storage system data representing a change in state of each component of the system over time;
A step of determining whether the processor is in a blocking state or a staging state based on system data read from the storage;
When it is determined that the idle component is in a blocking state, the processor replaces the idle component with another component that is adjacent to the idle component among the plurality of components. As a starting element from the starting element, and the tracked configuration based on data related to the tracked component of the system data read from the storage. Cause of determining whether an element is in an active state and causing the currently tracked component to block the idle component when it is determined that the component tracked this time is in an active state A first cause identifying step identified as an element;
When it is determined that the idle component is in the staging state, the processor selects another component of the plurality of components adjacent to the idle component upstream of the idle component. An element as a starting element is tracked from the starting element in the reverse direction of the flow, and is tracked based on data related to the tracked component of the system data retrieved from the storage. It is determined whether or not the component is in an active state, and when it is determined that the component tracked this time is in an active state, the component tracked this time is caused to cause staging of the idle component. A second cause identifying step that identifies the cause element,
Based on the system data read from the storage, the processor associates the cause element identified by the first and second cause identification processes with the cause element affecting the idle component. A route specifying step that specifies the cause route , and
In both the first and second cause identification steps, prior to the tracking, the processor ranks all adjacent components adjacent to the idle component according to a predetermined rule according to their respective states. And a ranking step of selecting any adjacent component as a component to be tracked according to the assigned ranking,
The route specifying step includes a step in which the processor specifies each cause route between the idle component and each cause element based on the rank assigned to each cause element. . Further, when the processor has at least one cause element for a plurality of idle periods of the same idle component based on the system data read from the storage, the processor sets the idle element for each idle element. The cause / path estimation method according to claim 1, further comprising an influence estimation step of estimating an influence on the component in association with each cause path from the cause element to the idle component. The ranking step includes the processor based on system data read from the storage when the idle component is in a blocking state; and
Rank 1: When the component is a blocked machine
Rank 2: When the component is a full buffer
Rank 3: When the component is a machine in an active state
Rank 4: When the component is a starved machine
Rank 5: If the component is a buffer that is not full
In accordance with the rule, including the step of assigning a rank to the target object as each adjacent component,
The first cause identifying step includes a step of tracking a plurality of adjacent components in the system from the object in a reverse direction of the flow when the order is given to the object. The cause / path estimation method according to Item 1 or 2 . The ranking step is based on system data read from the storage by the processor when the idle component is in a starving state, and
Rank 1: When the component is a starved machine
Rank 2: When the component is an empty buffer
Rank 3: When the component is a machine in an active state
Rank 4: If the component is a blocked machine
Rank 5: If the component is a non-empty buffer
In accordance with the rule, including the step of assigning a rank to the target object as each adjacent component,
The second cause identifying step includes a step of tracking a plurality of adjacent components in the system in the forward direction of the flow from the object when the order is given to the object. 4. The cause / route estimation method according to any one of Items 1 to 3 . Further, the processor maintains the state of the idle component by changing its buffer for each buffer position between components in the system based on system data read from the storage during the tracking. A statistical value including at least one of an amount of time change, a time during which each buffer continues to be the cause element, and a time during which each adjacent component adjacent to the idle component remains on the cause path The cause / path estimation method according to claim 1, further comprising a statistical value calculation step of calculating. A program executed by a computer to carry out the method according to any one of claims 1 to 5. The recording medium which recorded the program of Claim 6 so that computer reading was possible. In order to generate a flow from the upstream side to the downstream side of the processed element, a system in which a plurality of components to be processed with respect to the processed element are arranged according to a predetermined flow. A cause estimation device for estimating a cause element, which is another component that causes an idle state in a certain element, using a computer having a processor and a storage,
  Reading means for reading system data representing a change in state of each component in the system over time from the storage;
  Determining means for determining whether the idle component is in a blocking state or a staging state based on system data read from the storage;
  When it is determined that the idle component is in a blocking state, another component adjacent to the idle component on the downstream side of the plurality of components is used as the starting component. Tracking the forward direction of the flow from the starting element, and based on the data related to the tracked component of the system data read from the storage, the tracked component is activated If the component tracked this time is determined to be in an active state, the component tracked this time is identified as the cause factor that caused the blocking of the idle component First cause identification means;
  When it is determined that the idle component is in the staging state, another component adjacent to the idle component upstream of the idle component among the plurality of components is used as the starting component. As well as tracing from the starting element in the reverse direction of the flow Based on the data related to the tracked component among the system data read from the storage, it is determined whether or not the tracked component is in an active state. A second cause identifying means for identifying the component tracked this time as the cause element that caused the staging of the idle component when it is determined that the active component is in an active state;
  Including
  Both the first and second cause identifying means give priorities to all adjacent components adjacent to the idle component in accordance with a predetermined rule according to their respective states prior to the tracking, A cause estimation apparatus including ranking means for selecting any adjacent component as a component to be tracked according to the given ranking.

Images