JP7474152B2 - Production Planning System - Google Patents

Description

The present invention relates to a production planning system. In particular, it relates to a production planning system that creates production plans for a case in which multiple pieces of equipment operated by workers process multiple types of intermediate products one unit at a time.

At production sites where goods are produced, properly allocating production equipment and workers and increasing production efficiency is directly linked to lowering manufacturing costs and stabilizing management, and is an important management issue.

Known manufacturing methods include line production, which is suitable for small-variety mass production; cell production, which is suitable for large-variety small-volume production; and parallel process production, in which one worker operates and handles multiple pieces of equipment with the same function. In line production, which is suitable for small-variety mass production, intermediate products flow from the upper process to the lower process in sequence, and the production volume can be varied by adjusting the takt time, making it easy to plan production. In contrast, flexible cell production and parallel processes, which are suitable for large-variety small-volume production, produce multiple products in one manufacturing cell, and the products produced are frequently changed, so workers at the manufacturing site must frequently change the setup for this. Currently, in cell production and parallel process production, production plans are made based on the experience and intuition of the planning worker. For this reason, there is a demand for efficient and error-free control of manufacturing cells.

In order to solve such problems, Patent Document 1 describes a cell control system having a communication device that communicates with a manufacturing cell having multiple manufacturing machines for manufacturing products, an acquisition unit that acquires first manufacturing information for each manufacturing cell related to the manufacturing of the product, a first generation unit that generates multiple pieces of second manufacturing information for each of the multiple manufacturing machines based on the first manufacturing information, a transmission unit that transmits the multiple pieces of second manufacturing information to each of the multiple corresponding manufacturing machines via the communication device, a reception unit that receives third manufacturing information related to information for each manufacturing machine corresponding to the second manufacturing information from each of the multiple manufacturing machines via the communication device, a second generation unit that generates fourth manufacturing information for each manufacturing cell based on the multiple pieces of third manufacturing information from the multiple manufacturing machines, and an output unit that outputs the fourth manufacturing information.

JP 2017-33525 A

However, the invention described in Patent Document 1 is a line production system, and although it can create accurate production plans when the processing machines and robots always flow in the same order, when workers are involved, it is difficult for the production planning system to grasp the work of the workers. For this reason, there is a risk that the accuracy of the production plan will decrease in the case of cell production systems or parallel processes in which workers are involved.

In consideration of the above problems, the present invention aims to provide a production planning system that accurately grasps the operating status and creates accurate production plans in cell production methods with human intervention and parallel processes in which multiple pieces of equipment are operated.

The production planning system of the present invention, which aims to solve the above problem,

(1) A production planning system that creates a production plan for a process in which a plurality of pieces of equipment operated by workers process a plurality of intermediate products one unit at a time in a cell production system in which workers are involved or a parallel process in which a plurality of pieces of equipment are operated, the system comprising:
an input unit for inputting processing information and processing quantity for each type of the intermediate product;
an operating status analysis unit including: an operation detection sensor provided in each of the plurality of pieces of equipment, which determines a stop time and an operating time of the plurality of pieces of equipment; a replacement detection sensor which detects a replacement of the intermediate products in the plurality of pieces of equipment; and an operation detection sensor which detects whether the worker is working or waiting for work in the plurality of pieces of equipment;
a calculation unit which divides the downtime into pre-setup time, post-setup time, and work waiting time based on a signal from the operation status analysis unit when the processing of the same intermediate product is repeated, obtains a standard processing time indicating the processing time of the intermediate product for each type of product by using the pre-setup time, the operation time, and the post-setup time as one unit, and adjusts between the plurality of facilities to allocate the processing quantities;
an output unit that issues processing instructions to each piece of equipment based on the allocation result of the calculation unit;
The production planning system is characterized by having the following features.

According to the present invention, the equipment stop time detected by the operation detection sensor is divided into pre-setup time, post-setup time, and waiting time for work by the worker, and the pre-setup work, equipment operation, and post-setup work are treated as one unit and the standard processing time required for them is calculated. This eliminates waiting time and obtains a highly accurate standard processing time for each processing information, making it possible to appropriately allocate intermediate products to be processed.

The present invention also preferably has the following aspects:

(2) The calculation unit of the production planning system further extracts a product type changeover time from the stop time to obtain a standard product type changeover time when the product type of the intermediate product being processed is changed, and adjusts and allocates the processing quantity between the multiple pieces of equipment based on the standard processing time and the standard product type changeover time.

(3) The operation detection sensor is a sensor that detects the lighting of an indicator light provided on the equipment.

(4) The work detection sensor is a human presence sensor that is provided in front of the equipment and determines whether the worker is working.

(5) The work detection sensor is a safety device sensor that is provided on the equipment and determines whether the worker is working.

(6) The replacement detection sensor is a reader that reads the replacement timing of the identification data sent to the process together with the intermediate product or the processing data sent to the equipment.

(7) The operation status analysis unit has a worker identification unit that identifies the worker, and the calculation unit calculates the standard processing time for each worker.

(8) The production planning system repeatedly reallocates the remaining intermediate products to be processed based on production results.

(9) The calculation unit has a simulator that predicts the production quantity based on information from the input unit and the operation status analysis unit, and the calculation unit allocates intermediate products to be processed based on the results of the simulator.

(10) The calculation unit has a learning unit that performs reinforcement learning.

(11) The calculation unit performs reinforcement learning using Q-learning, in which the reward is the time required to complete the processing quantity.

(12) The equipment includes one or more devices selected from the group consisting of processing machines, robots, PLCs, conveyors, measuring instruments, test equipment, presses, presses, printing machines, die-casting machines, injection molding machines, food processing machines, packaging machines, welding machines, cleaning machines, painting machines, assembly machines, mounting machines, woodworking machines, sealing devices, and cutting machines.

According to the present invention, the equipment stop time detected by the operation detection sensor is divided into pre-setup time, post-setup time, and waiting time for work by the worker, and the pre-setup time, operation time, and post-setup time are used as one unit to calculate the standard processing time, thereby obtaining a highly accurate standard processing time for each processing information item, with the waiting time eliminated. As a result, the intermediate products to be processed can be appropriately allocated, improving work efficiency and enabling more products to be produced in a shorter time. This can contribute to improving equipment operating rates and labor productivity.

In addition, according to the present invention, even in cases where the intermediate product only needs to be processed within a certain time frame, such as in the case of a process prior to a batch process, and workers have a great deal of freedom, it is possible to obtain appropriate production information and make efficient production plans.

Furthermore, the calculation unit of the production planning system obtains the standard product type changeover time when the product type of the intermediate product being processed is changed, and allocates the processing quantity to the equipment based on the standard processing time and the standard product type changeover time, thereby improving the accuracy of allocation, especially when the work time associated with product type changeover is long, such as changing dies and blades, returning to the starting point, and cleaning.

By using an operation detection sensor that detects the illumination of an indicator light installed on the equipment, operation information can be easily obtained from the equipment without any special modifications.

By using a human presence sensor as the task detection sensor, it is easy to determine which device the worker is working on.

By using the operation detection sensor as a safety device sensor, when an operator releases the safety device and is performing pre-setup, post-setup, or product type changeover, it is possible to easily determine which equipment the operator is working on.

By making the changeover detection sensor a reader that reads the identification data sent to the process together with the intermediate product, or the timing of changing the processing data sent to the equipment, when work instructions are sent together with the intermediate product, it is possible to read the identification code written on the work instructions to identify which intermediate product is being processed. Also, when the processing program for the intermediate product to be processed is sent over the network, it is possible to determine that the type is being changed based on the writing of the processing program.

The operation status analysis unit has a worker identification unit that identifies the worker, and the calculation unit calculates the standard processing time for each worker, so that when there are multiple workers, the pre-setup time, post-setup time, and product type changeover time for each worker can be determined by identifying the workers. Furthermore, when there are differences in work efficiency between workers, the accuracy of allocation can be improved.

The production planning system repeatedly reallocates the remaining intermediate products to be processed in the future based on production results, allowing reallocation not only at the start of equipment operation but also during the process, making it possible to appropriately revise the production plan even in the event of a sudden breakdown or interruption.

The calculation unit has a simulator that predicts production quantities based on information from the input unit and the operation status analysis unit, and the calculation unit allocates intermediate products to be processed based on the results of the simulator. By having the calculation unit have a simulator, it is possible to create highly accurate production plans using simple algorithms. By having the calculation unit have a learning unit that performs reinforcement learning, it is possible to increase the speed of calculations.

The calculation unit performs reinforcement learning using Q-learning, which uses the time required to complete a processing quantity as a reward, and by using the time required to complete a processing quantity as a reward in Q-learning, it is possible to efficiently allocate tasks to complete the processing time more quickly.

When the equipment includes one or more devices selected from the group consisting of processing machines, robots, PLCs, conveyors, measuring instruments, test equipment, presses, presses, printing machines, die-casting machines, injection molding machines, food machinery, packaging machines, welding machines, cleaning machines, painting machines, assembly equipment, mounting machines, woodworking machines, sealing equipment, and cutting machines, by applying a production planning system to these devices, it is possible to efficiently and appropriately create production plans for processes using the equipment.

FIG. 1 is a diagram illustrating a schematic configuration of a production planning system according to a first embodiment of the present invention. FIG. 2 is a diagram illustrating, in the form of a time chart, the processes executed by the operation status analyzer and the simulator in the first embodiment. FIG. 3 shows an algorithm for determining which machine operation the worker should give priority to in the first embodiment. FIG. 4 is a schematic diagram showing a method for calculating the number of combinations for allocating intermediate products to each processing machine. FIG. 5 is an explanatory diagram showing how learning proceeds in the second embodiment according to the present invention.

Hereinafter, an embodiment of the present invention will be described.
The production planning system of the present invention is a production planning system that makes a production plan for a process in which a plurality of pieces of intermediate products are processed one unit at a time by a plurality of pieces of equipment operated by workers,
an input section for inputting processing information and processing quantity for each type of intermediate product;
an operating status analysis unit having an operation detection sensor provided in the equipment for determining the stop time and operation time of the plurality of pieces of equipment, a replacement detection sensor for detecting replacement of intermediate products in the plurality of pieces of equipment, and an operation detection sensor for detecting whether an operator is working or waiting for work in the plurality of pieces of equipment;
a calculation unit which divides downtime into pre-setup time, post-setup time, and work waiting time based on a signal from the operation status analysis unit when the processing of the same intermediate product is repeated, obtains a standard processing time indicating the processing time of the intermediate product for each type, with the pre-setup time, operation time, and post-setup time being taken as one unit, and adjusts between the plurality of facilities to allocate the processing quantities;
an output unit that issues processing instructions to each piece of equipment based on the allocation result of the calculation unit;
has.

In the present invention, "a process in which multiple pieces of equipment operated by workers process intermediate products one unit at a time" refers to a process in which one unit of standard work is made up of pre-setup work, automatic processing of the intermediate products by equipment, and post-setup work. The time required to complete one unit of standard work is called the standard processing time. Here, pre-setup work is work in which an operator manually sets an object, which is an intermediate product, in the equipment while the equipment is stopped. Also, post-setup work is work in which an operator manually removes the intermediate product from the equipment while the equipment is stopped. The time it takes to perform one pre-setup work is the pre-setup time, and the time it takes to perform one post-setup work is the post-setup time.

For the purpose of more specifically explaining the embodiment of the production planning system of the present invention, in the drawings and the following description, each of the multiple facilities M is identified by the reference characters _M1 , _M2 , _M3 , ... _Mm . Furthermore, the multiple types of intermediate products P are identified by the reference characters _P1 , _P2 , _P3 , ... _Pn for each type, and the processing quantities determined for each type of intermediate product P are represented by the reference characters _Q1 , _Q2 , _Q3 , ... _Qn . In other words, the processing quantities for each type are associated with each other by being given the same subscript, as shown in Table 1.

In the process of creating _a production plan, the production planning system sets an optimal processing sequence for each piece of equipment ( _M1 , _M2 , _M3 , ... _Mm ) so that the total time required to manufacture a predetermined number of pieces (Q1, _Q2 , _Q3 , ... _Qn ) of each of a plurality of varieties ( _P1 , _P2 , P3 ... _Pn ) using a plurality of pieces of equipment ( _M1 , _M2 , M3 ... _Mm ) is minimized. Specifically, a production plan is created for each piece of equipment, for example, in the order _P1 , _P1 , P1, _P2 , _P2 ... for _M1 , and _P3 , _P3 , _P2 , _P2 , _P2 ... for M2.

In the following, the total amount of processed intermediate products is represented as Q _total (=Q ₁ +Q ₂ +Q ₃ . . . +Q _n ).

The pre-setup time, operation time, and post-setup time calculated by the operation status analysis unit are obtained by detecting one actual unit of work process for each product type using a sensor and measuring the time required for each task. In this specification, variables are set for the time corresponding to each task for each product type, as shown in Table 2. Note that the subscripts before and after the product type changeover time correspond to the product type before and after the changeover, respectively. Also, the operation time is sometimes specifically described as "net processing time," meaning the net time during which the equipment is automatically operating and actually processing the intermediate product.

The pre-setup time for product type Pk is indicated by Tprek, the net processing time by Tnetk, and the post-setup time by Tpostk, and the product switching time required to switch from product type i to product type j is indicated by Tswij. Note that when i=j (when product switching does not occur), Tswij=0.

In the present invention, in order to obtain an accurate value of the standard processing time, which is the sum of pre-setup time, operation time, and post-setup time, it is preferable to accumulate multiple data and take an average. Similarly, in order to obtain an accurate value of the standard product type switching time, it is preferable to accumulate switching data and improve accuracy.

For both pre-setup and post-setup time, only the time the worker is working on the equipment in question is counted, and the time spent working on other equipment is not counted. Pre-setup and post-setup time during equipment downtime can be detected by a work detection sensor that detects the equipment the worker is working on.

The product type changeover time is the new work time required for an operator to process a different product type on equipment after one unit of standard work is completed. The product type changeover time can be determined by subtracting the normal post-setup time and pre-setup time from the time when the equipment is stopped and the operator is working on the stopped equipment. The standard product type changeover time can be obtained by measuring the product type changeover time multiple times and taking the average value. Alternatively, the standard time can be accurately measured by simulating only the product type changeover work while the equipment is stopped.
In addition, when processing intermediate products that do not require a product type changeover time, the present invention can be applied without taking the product type changeover time into consideration.

Work waiting time is the time during which the operation detection sensor detects that the equipment is idle, and the work detection sensor detects that no worker is working on the equipment. In other words, it refers to the time during which the equipment is idle, even though it is operational.

The pre-setup time, operation time, and post-setup time for each type may be obtained each time a production plan is created that assigns the type and processing quantity of intermediate products to each piece of equipment. Alternatively, the data may be obtained over a long period of time and the values may be stored in an internal or external storage device. Using long-term data can increase the reliability of the data.

The standard processing time may be obtained by identifying the worker, or it may not be necessary to identify the worker. If differences in proficiency between workers affect the standard processing time, it is desirable to identify them as this will become a source of error, but if the effect is small, it will be easier to create a production plan if data is not obtained for each worker.

Below, we will explain in more detail, with reference to the drawings, embodiments 1 and 2 of the production planning system of the present invention, which creates production plans using data on standard processing times and standard product type changeover times.

<First embodiment>
Fig. 1 is a diagram showing a schematic configuration of a production planning system according to the present embodiment 1. The production planning system 1 according to the present embodiment 1 will be described with reference to Fig. 1.

The production planning system 1 of the first embodiment is an example of a production planning system that allocates an optimal processing sequence for intermediate products when one worker 100 processes multiple types of intermediate products by operating three pieces of equipment _M1 , _M2 , and _M3 . The equipment _M1 , _M2 , and _M3 in this embodiment are processing machines having the same performance and capable of processing one intermediate product at a time.

In this embodiment, the unprocessed intermediate products 51 are stored one by one on a tray 61 together with a processing instruction sheet 41. A day's worth of processing quantity of unprocessed intermediate products 51 is prepared. Similarly to the unprocessed intermediate products, the processed intermediate products are stored one by one on a tray together with a processing instruction sheet. When a day's worth of processing quantity of processed intermediate products has been collected, they are combined into one lot and sent to the next batch process. For this reason, there are no particular restrictions on the processing order of the intermediate products, and they can be processed in any order.

In this embodiment, when the intermediate products 51 are processed, the intermediate products 51 are placed on a dedicated table provided for each piece of equipment together with the tray 61. The dedicated table is equipped with a replacement detection sensor 3, which detects which intermediate product 51 is being processed. The replacement detection sensor 3 can be, for example, a reader for the barcode or two-dimensional code on the processing instruction sheet 41, or a communication device capable of communicating with the wireless tag on the tray 61. After processing, the intermediate products 51 are returned to the tray 61 and replaced together with the tray 61, so that it can be determined that the intermediate products 51 have been replaced when the tray 61 is replaced.

The configuration of the replacement detection sensor 3 is not particularly limited. It is not limited to a code reading device or a device for communicating with a wireless tag, but may be an image recognition device for the intermediate product 51 set in the equipment, or any other device may be applied.

In front of each of the three pieces of equipment _M1 , _M2 , and _M3 , a pressure sensor is laid as a work detection sensor 4. The pressure sensor outputs a signal only when the worker 100 stands on the equipment _M1 , M2, and _M3 while the equipment M1, _M2 , and M3 are stopped, and determines which piece of equipment the worker 100 is working on. If the worker 100 stands on the equipment for a short time, the work detection sensor 4 does not output a signal by filtering. The work detection sensor 4 is not limited to a pressure sensor, and any sensor that detects the presence or movement of the worker 100, such as an infrared sensor, a radar, or a range finder, can be used. In addition, a camera and an image processing device that constantly takes pictures with a camera and uses image recognition to determine which piece of equipment the worker 100 is working on, can also be used as the work detection sensor 4.

Each of the three pieces of equipment _M1 , _M2 , and _M3 is equipped with an indicator light 22 that indicates whether the equipment is running or stopped. An operation detection sensor 21 consisting of an optical sensor is arranged next to each indicator light 22 so as to be paired with the indicator light 22, and detects the operation state of the equipment. The form of the operation detection sensor 21 is not limited to an optical sensor. For example, the power supply of the indicator light may be divided to provide an appropriate voltage, and the voltage value may be input directly to the operation status analysis units 11, 12, and 13 of the respective devices to function as the operation detection sensor 21.

In this embodiment, PLCs (Programmable Logic Controllers) connected to the equipment _M1 , _M2 , and _M3 function as operation status analysis units 11, 12, and 13 that aggregate signals from sensors of each equipment. The operation status analysis units 11, 12, and 13 wirelessly send signals indicating the operation status of the equipment _M1 , _M2 , and M3 to a computer ₂ that serves as a calculation unit.

Because the computer 2 and the PLC are operated as an integrated unit, both the computer 2 and the PLC can be configured to handle distributed functions of the operation status analysis unit. Also, the computer 2 may function as both the calculation unit and the operation status analysis unit, and each sensor may be directly connected to the computer 2 to perform the required processing.

A touch panel 5 that serves as both an output unit and an input unit is connected to the computer 2. The type and quantity of intermediate products to be processed that day are input from the touch panel 5. Note that the input unit is not limited to one that accepts manual input, but may be a communication cable that inputs signals, or an interface board that sends digital information wirelessly to the calculation unit.

The touch panel 5 also sequentially displays the processing order of the intermediate products 51 in each piece of equipment, which is obtained by the calculation process performed by the calculation unit of the computer 2. The intermediate products 51 that will be processed in the nearest step by each piece of equipment are displayed on the monitors 42 connected to the PLC of each piece of equipment _M1 , _M2 , _M3 , so that the worker 100 can accurately receive instructions on the intermediate products 51 to be processed next. Each piece of equipment _M1 , _M2 , _M3 may also be provided with a mechanism for sequentially sending out trays 61 required for the next job according to a command from the PLC.

Next, a process executed by the simulator of the production planning system of this embodiment to allocate the intermediate product 51 to each of the facilities M ₁ , M ₂ , and M ₃ and issue a command to the PLC will be described.

In this embodiment, the production plan is not instructed by machine learning, but rather, the simulator in the calculation unit runs simulations for all combinations according to certain rules, selects the processing order that will minimize the processing time, and sends instructions.

The simulation process performed by the simulator of this embodiment will now be described. Figure 3 is a flowchart showing an example of an algorithm executed by the simulator to instruct the actions of a worker.

In this algorithm, when the worker 100 is free, the work is assigned so that the next equipment is operated as soon as possible. When the worker is free, step S1 is started. In the next step S2, it is determined whether there is one equipment waiting for work. If there is one equipment waiting for work, step S2 becomes NO, and the process moves to step S5, where a series of work is continuously performed, including post-setup of the intermediate product 51 set in the corresponding equipment, product type switching work, pre-setup of the next intermediate product 51, and start of processing. If there are multiple pieces of equipment waiting for work, step S2 becomes YES, and the process moves to step S3. Here, the time required to start the equipment waiting for work is calculated. Next, in step S4, the equipment that can start processing in the shortest time is selected. Then, in step S5, a series of work is continuously performed, including post-setup of the intermediate product 51 set in the corresponding equipment, product type switching work, pre-setup of the next intermediate product 51, and start of processing.

According to the above algorithm, the simulation is repeated until the processing of all the intermediate products 51 is completed, and the pattern that can complete the processing of the planned processing quantity in the shortest time is found from among all the patterns, so that the processing sequence of the intermediate products 51 can be appropriately allocated to each of the facilities _M1 , _M2 , and _M3 .

The outline of the calculations that the simulator repeatedly performs to allocate intermediate products 51 to equipment for processing will be described with reference to Fig. 4. In Fig. 4, among the intermediate products 51 before processing, product type _P1 is shown by a circle, product type _P2 is shown by a square, and product type _P3 is shown by a triangle (Fig. 4(A)).

In the following, as an example for a simpler explanation, a case will be described in which a total of ten intermediate products 51 are assigned to three machines _M1 , _M2 , and _M3 of the same specifications for processing. Here, the ten intermediate products 51 are broken down into three products of type _P1 ( _Q1 =3), four products of type _P2 ( _Q2 =4), and three products of type _P3 ( _Q3 =3).

When a total of 10 intermediate products 51 consisting of types _P1 , _P2 , and _P3 are allocated to three pieces of equipment _M1 , _M2 , and _M3 , the number of all combinations is equal to the number of combinations when 10 pieces _P1 , _P2 , and _P3 are lined up in a row and cut at two points in the middle of the row to divide it into three groups, as shown in FIG. 4(B).

In other words, when 10 intermediate products 51 of three kinds are arranged in a row, the number of combinations is as follows:
(3+4+3)!/(3!×4!×3!)=4200 possibilities. By considering this string as being cut in two places and divided into three sets, the intermediate products 51 can be allocated to the three pieces of equipment M ₁ , M ₂ , and M ₃ .

The number of ways to cut a row of 10 intermediate products in two places is 11+10+9+8+7+6+5+4+3+2+1=66. This includes the case where the end of the row is cut to form a set of 0 products and divide the intermediate products 51 into two or one group. 4200×66=277,200 ways is the total number of patterns when an intermediate product 51 consisting of _Q1 =3, _Q2 =4, and _Q3 =3 is processed by three pieces of equipment _M1 , _M2 , and _M3 (FIG. 4C).

For all combinations of patterns, the earliest possible processing order can be assigned using pre-setup time, pre-processing time, post-setup time, and variety switching time that differs depending on the variety before and after the switchover, and calculations using a simulator that sets the work waiting time for times when the work of workers 100 overlaps.

As an example of one of the 277,200 simulations of this embodiment, a method for determining a production plan for each piece of equipment when the following processing sequence is adopted will be described with reference to the time chart of FIG. 2 and the algorithm of FIG. 3.
The processing sequence of the intermediate products by the equipment _M1 is _P1 → _P1 → _P3 .
The processing sequence of the intermediate products by the equipment _M2 is _P2 → _P2 → _P1 → _P2 .
The processing sequence of the intermediate products by the equipment _M3 is _P3 → _P3 → _P2 .

At the time ( _t0 ) when the equipment starts operating, the worker 100 is free and the status is step S1. At this time, the three pieces of equipment _M1 , _M2 , and _M3 are waiting for processing, and step S2 is YES. The simulator calculates the time until each piece of equipment starts operating (step S3), selects the equipment _M3 that processes the product type _P3 that requires the shortest pre-setup (step S4), and determines that the worker 100 should start pre-setup work (step S5).
At the time ( _t1 ) when the pre-setup of equipment _M3 is completed, the worker 100 is free again and the state of step S1 is entered. At this time, equipment _M1 and equipment _M2 are waiting for processing, and step S2 is YES. The simulator calculates the time until equipment _M1 and equipment _M2 start operating (step S3), selects equipment _M2 to process _P2 , which has a shorter pre-setup time (step S4), and determines that the worker 100 will perform the pre-setup work.
At the time ( _t2 ) when the pre-setup of equipment _M2 is completed, the worker 100 becomes available again. However, at this time, only equipment _M1 is waiting for processing, so step S2 becomes NO, the simulator process moves to step S5, and it is determined that the worker 100 will perform the pre-setup work of equipment _M1 .
After that, when all three pieces of equipment _M1 , _M2 , and _M3 are in operation, the worker 100 waits with his hands free. After that, the operation time of equipment _M2 ends earliest ( _t4 ), so step S2 becomes NO and the simulator determines that the worker 100 should start post-setup work for equipment _M2 and the subsequent pre-setup work for product _P2 .

In this way, the simulator allocates one unit of standard work and product type changeover work to the three pieces of equipment _M1 , _M2 , and _M3 by the worker 100 according to the algorithm shown in Fig. 3. Then, it is possible to calculate the time from the start of operation of the equipment until the processing of a specified number of intermediate products 51 is completed.

By performing the above simulations for all patterns, it is possible to determine the processing sequence that will result in the shortest processing time and create a production plan. In addition, even during actual processing work, simulations can be performed at any time using the remaining number of items and the operating status of the equipment, and the production plan can be changed as appropriate.

In the simulation method of this embodiment, the number of patterns increases exponentially as the number of intermediate products 51 to be processed increases, and the load on the calculation unit increases. For example, when the number of intermediate products 51 is doubled to a total of 20, and allocation is performed with _Q1 = 6, _Q2 = 8, and _Q3 = 6, the total number of patterns is 7,682,154,480. When the number of intermediate products 51 is doubled to a total of 100, and allocation is performed with _Q1 = 30, _Q2 = 40, and _Q3 = 30, the total number of patterns is approximately 1.07 x ^1047. The production planning system of this embodiment is particularly suitable for creating a production plan when there is an upper limit on the variety and the number of intermediate products 51 to be processed.

<Embodiment 2>
In this embodiment, a system for efficiently allocating a processing sequence and creating a production plan will be described even when there are a large number of types and intermediate products 51 to be processed.

In this embodiment, the same types of intermediate products are assigned to the same equipment as in embodiment 1. The difference from embodiment 1 is that the simulator of the calculation unit assigns the optimal processing order by machine learning using the Q-learning method. Components that have the same configuration as embodiment 1 are given the same reference numerals and their explanations are omitted.

Q-learning is a learning method that selects an action a to respond to a situation s in a specific environmental state s, calculates a value function Q(s, a) that indicates the value of action a, and selects the action a that maximizes the value function Q(s, a) as the optimal action, repeatedly deriving a more desirable conclusion.

The method of Q-learning will be described below. The value function Q is calculated according to the following formula (1), and the obtained value is called the Q value.
Q(s, a) ← Q(s, a) + αr (1)
Here, r is the reward.
s: Status
a: Action selected in state s
α: learning coefficient

The reward r shown in formula (1) is an index for evaluating action a in response to a change in state s caused by action a. The reward r can be set in advance. Typically, reward r is set to a positive value so that the Q value increases if action a causes a favorable change in state s, and a negative value so that the Q value decreases if action a results in an unfavorable change. In this embodiment, learning can be progressed by setting reward r to 1 if the total time required for processing becomes shorter after repeated trials than the previous time, setting reward r to 0 if the value remains the same, and setting reward r to -1 if the time becomes longer.

In this embodiment, the state s is defined as:
s ₁ : raw number of P ₁ ;
_s2 : raw number of _P2 ;
_s3 : raw number of _P3 ;
_s4 : The variety that _M1 had been processing until just before,
_s5 : The variety that _M2 had been processing until just before,
_s6 : The variety that _M3 had been processing until just before;
s ₇ : Equipment waiting for work,
This results in a seventh-order array variable.
The action a selected in state s is
a ₁ : Equipment for starting the next processing;
_a2 : the variety to start the next processing;
It becomes a second-order array variable.
Regarding the equipment waiting for work in _s7 , if the equipment M _i is in operation, it is represented as "1", and if it is waiting for work, it is represented as "0". For example, if M ₃ is in operation, M ₂ is waiting for work, and M ₁ is waiting for work,
_s7 =4(0b100)
It can be expressed as a 3-bit binary number.

For simplification, only _s1 , _s2 , and _s3 are extracted, and the state of learning the progress of machining using a three-dimensional diagram is shown diagrammatically in Fig. 5. That is, Fig. 5 shows an array of 3x4x3 ( _s1 , _s2 , _s3 ), but in actual learning, learning is performed based on an array of 3x4x3x3x3x3x3 states by adding the state ( _s4 , _s5 , _s6 , _s7 ) hidden in Fig. 5.

5 shows multiple routes starting from the origin and leading to the processing quantity of a predetermined intermediate product, ( _s1 , _s2 , _s3 ) = (3, 4, 3), with arrows of different line types for each piece of equipment. Processing by _M1 is shown with a solid line, processing by _M2 with a dashed line, and processing by _M3 with a double line.

In Q-learning, machine learning randomly assigns a Q value, and intermediate products are selected and assigned, and then repeated trials are performed until convergence occurs.
A specific method for the n-th learning will be described below, but the method for applying formula (1) is not limited to this.

At the start time _t0 ,
Since the processed numbers of _P1 , _P2 , and _P3 are 0 (0,0,0), the unprocessed numbers are ( _s1 , _s2 , _s3 ) = (3,4,3). Since this is the start, _M1 is initially set to _P1 , _M2 to _P2 , and _M3 to _P3 , and ( _s4 , _s5 , s6) = (1,2,3).
Since each piece of equipment starts in a waiting state for work, ( _M3 , _M2 , _M1 ) = (0, 0, 0), i.e., _s7 = 0 ( 0b000 ), the state can be expressed as ( _s1 , _s2 , _s3 , _s4 , _s5 , _s6 , _s7 ) = (3, 4, 3, 1, 2, 3, 0).
In addition, action a can be performed for three types of products on three pieces of equipment, and the action with the highest value is selected from the nine (3 x 3) Q values that are assigned in advance. (Here, _M3 starts processing _P3 , and _a1 = 3, _a2 = 3)

At _t1 when the first pre-setup work is completed,
The remaining number of _P3 decreases, the processed numbers of _P1 , _P2 , and _P3 become (0, 0, 1), and the unprocessed numbers become ( _s1 , _s2 , _s3 ) = (3, 4, 2).
There is no change in s ₄ to s ₆ , so (s ₄ , s ₅ , s ₆ )=(1, 2, 3).
For _s7 , the waiting tasks are _M1 and _M2 , and ( _M3 , _M2 , _M1 ) = (1, 0, 0), i.e., _s7 = 4 ( 0b100 ).
( _s1 , _s2 , _s3 , _s4 , _s5 , _s6 , _s7 ) = (3, 4, 2, 1, 2, 3, 4)
The state can be expressed as:
In addition, action a can be performed for three varieties on two machines, and the action with the highest Q value is selected from the six (2 x 3) Q values that are assigned in advance. (Here, _M2 starts processing _P2 , and _a1 = 2 and _a2 = 2 are selected.)

Next, at time _t2 ,
The remaining number of _P2 decreases, and the processed number is ( _P1 , _P2 , _P3 ) = (0, 1, 1), so the unprocessed number is ( _s1 , _s2 , _s3 ) = (3, 3, 2).
There is no change in s ₄ to s ₆ , so (s ₄ , s ₅ , s ₆ )=(1, 2, 3).
For _s7 , the waiting work is _M1 , and ( _M3 , _M2 , _M1 ) = (1, 1, 0), that is, _s7 = 6 ( 0b110 ),
( _s1 , _s2 , _s3 , _s4 , _s5 , _s6 , _s7 ) = (3, 3, 2, 1, 2, 3, 6)
The state can be expressed as:
In addition, action a can be performed on one piece of equipment for three different types of products, and the action with the highest Q value is selected from the three (1 x 3) Q values that are assigned in advance. (Here, _M1 starts processing _P1 , and selects _a1 = 1 and _a2 = 1.)

At time _t4 ,
The remaining number of _P1 decreases, the number of processed pieces becomes ( _P1 , _P2 , _P3 ) = (1, 1, 1), and the number of unprocessed pieces becomes ( _s1 , _s2 , _s3 ) = (2, 3, 2).
There is no change in s ₄ to s ₆ , so (s ₄ , s ₅ , s ₆ )=(1, 2, 3).
For _s7 , the waiting work is _M2 , and ( _M3 , _M2 , _M1 ) = (1, 0, 1), i.e., _s7 = 5 ( 0b101 ),
( _s1 , _s2 , _s3 , _s4 , _s5 , _s6 , _s7 ) = (2, 3, 2, 1, 2, 3, 5)
The state can be expressed as:
In addition, action a can handle three varieties per machine, and the action with the highest Q value is selected from the three (1 x 3) Q values that are assigned in advance. (Here, _M2 starts processing _P2 , and selects _a1 = 2 and _a2 = 2.)

At time _t7 ,
The remaining number of _P2 decreases, the number of processed pieces becomes ( _P1 , _P2 , _P3 ) = (1, 2, 1), and the number of unprocessed pieces becomes ( _s1 , _s2 , _s3 ) = (2, 2, 2).
There is no change in s ₄ to s ₆ , so (s ₄ , s ₅ , s ₆ )=(1, 2, 3).
For _s7 , the waiting work is _M1 , and ( _M3 , _M2 , _M1 ) = (1, 1, 0), that is, _s7 = 6 ( 0b110 ),
( _s1 , _s2 , _s3 , _s4 , _s5 , _s6 , _{s7) = (2, 2, 2, 1, 2, 3, 6} )
The state can be expressed as:
In addition, action a can be performed on one piece of equipment for three different types, and the action with the highest Q value is selected from the three (1 x 3) pre-allocated Q values. (Here, _M1 starts processing _P1 , and selects _a1 = 1 and _a2 = 1.)

At time _t10 ,
The remaining number of _P1 decreases, the number of processed pieces becomes ( _P1 , _P2 , _P3 ) = (2, 2, 1), and the number of unprocessed pieces becomes ( _s1 , _s2 , _s3 ) = (1, 2, 2).
There is no change in s ₄ to s ₆ , so (s ₄ , s ₅ , s ₆ )=(1, 2, 3).
For _s7 , the waiting work is _M3 , ( _M3 , _M2 , _M1 ) = (0, 1, 1), that is, _s7 = 3 ( 0b011 ),
(s1, _s2 , _s3 , _s4 , _s5 , _s6 , _s7 ) = ( ₁ , 2, 2, 1, 2, 3, 3)
The state can be expressed as:
In addition, action a can handle three different types of products per machine, and the action with the highest Q value is selected from the three (1 x 3) Q values that are assigned in advance. (Here, _M3 starts processing _P3 , and selects _a1 = 3, _a2 = 3.)

At time _t13 ,
The remaining number of _P3 decreases, the number of processed pieces becomes ( _P1 , _P2 , _P3 ) = (2, 2, 2), and the number of unprocessed pieces becomes ( _s1 , _s2 , _s3 ) = (1, 2, 1).
There is no change in s ₄ to s ₆ , so (s ₄ , s ₅ , s ₆ )=(1, 2, 3).
For _s7 , the waiting work is _M2 , and ( _M3 , _M2 , _M1 ) = (1, 0, 1), i.e., _s7 = 5 ( 0b101 ),
(s1, _s2 , _s3 , _s4 , _s5 , _s6 , s7) = ( ₁ , 2, 1, 1, 2, 3, ₅ )
The state can be expressed as:
In addition, action a can handle three different types of products per machine, and the action with the highest Q value is selected from the three (1 x 3) Q values that are assigned in advance. (Here, _M2 starts processing _P1 , and selects _a1 = 2 and _a2 = 1.)

At time _t17 ,
The remaining number of _P1 decreases, the number of processed pieces becomes ( _P1 , _P2 , _P3 ) = (3, 2, 2), and the number of unprocessed pieces becomes ( _s1 , _s2 , _s3 ) = (0, 2, 1).
Since the variety processed by _M2 has changed to _P1 , ( _s4 , _s5 , _s6 ) = (1, 1, 3).
For _s7 , the waiting work is _M1 , and ( _M3 , _M2 , _M1 ) = (1, 1, 0), that is, _s7 = 6 ( 0b110 ),
( _s1 , _s2 , _s3 , _s4 , s5, _s6 , s7) = (0, 2, 1, ₁ , 1, 3, ₆ )
The state can be expressed as:
In addition, action a can handle three different types of products per machine, and the action with the highest Q value is selected from the three (1 x 3) Q values that are assigned in advance. (Here, _M1 starts processing _P3 , and selects _a1 = 1 and _a2 = 3.)

At time _t21 ,
The remaining number of _P3 decreases, the number of processed pieces becomes ( _P1 , _P2 , _P3 ) = (3, 2, 3), and the number of unprocessed pieces becomes ( _s1 , _s2 , _s3 ) = (0, 2, 0).
Since the variety of product processed by _M1 has changed to _P3 , ( _s4 , _s5 , _s6 ) = (3, 1, 3).
For _s7 , the waiting work is _M3 , ( _M3 , _M2 , _M1 ) = (0, 1, 1), that is, _s7 = 3 ( 0b011 ),
( _s1 , _s2 , _s3 , _s4 , _s5 , _s6 , _s7 ) = (0, 2, 0, 3, 1, 3, 3)
The state can be expressed as:
In addition, action a can handle three different types of products per machine, and the action with the highest Q value is selected from the three (1 x 3) pre-allocated Q values. (Here, _M3 starts processing _P2 , and selects _a1 = 3 and _a2 = 2.)

At time _t25 ,
The remaining number of _P2 decreases, the number of processed pieces becomes ( _P1 , _P2 , _P3 ) = (3, 3, 3), and the number of unprocessed pieces becomes ( _s1 , _s2 , _s3 ) = (0, 1, 0).
Since the product type processed by _M3 has changed to _P2 , ( _s4 , _s5 , _s6 ) = (3, 1, 2)
For _s7 , the waiting work is _M2 , and ( _M3 , _M2 , _M1 ) = (1, 0, 1), i.e., _s7 = 5 ( 0b101 ),
( _s1 , _s2 , _s3 , _s4 , _s5 , _s6 , _s7 ) = (0, 1, 0, 3, 1, 2, 5)
The state can be expressed as:
In addition, action a can handle three different types of products per machine, and the action with the highest Q value is selected from the three (1 x 3) pre-allocated Q values. (Here, _M2 starts processing _P2 , and _a1 = 2, _a2 = 2 are selected.)

At time _t34 , all processing is completed, and the number of unprocessed pieces is ( _s1 , _s2 , _s3 ) = (0, 0, 0).
Thus, the total processing time required to process all the intermediate products is obtained as t ₃₄ .

At the start time of each pre-setup task, the selection of which equipment to process which product is made by selecting the action that results in the largest Q value for each "action a" based on the "state s" at that time, and the total processing time for the nth trial is obtained.

If the obtained total machining time is smaller than that of the (n-1)th trial, the Q value of each passed point is decreased by α, and if it is larger, α is added to it. If they are the same, the value is not changed.
The Q value in each state is initially randomly assigned and then optimized through repeated trials.

In the first trial, the calculation unit randomly assigns a Q value to each of the seven states (s ₁ to s ₇ ), and the simulator calculates the time according to the instructions given by the machine learning while completing the allocation of all planned intermediate products according to the algorithm shown in Figure 3. In the first trial (n=1), the initial value of the total processing time is obtained. Since the reward in the first trial is 0, there is no change in the Q value for each action a that assigns one intermediate product that has passed through.

In the second trial, some parts are intentionally set aside to change the route from the first time regardless of the size of the Q value, and for the rest, the larger Q value given to action a, which is one unit of standard work for the selected equipment and product type, is selected, aiming to complete the production plan. For this reason, the total processing time will not be the same as the first time (n=1), and it is possible to judge whether the second trial is appropriate or not.

From the third time onwards, the aim is to complete the production plan by selecting the larger Q value given to action a, which is one unit of standard work for the selected equipment and product type.

By performing repeated trials, the allocation of the machining order that shortens the machining time will converge. It is preferable to determine the appropriate number of trials while monitoring the reduction in machining time for each trial, and use this as the output of machine learning.

Here, if the reward is 0, such as in the second trial where the decision is based solely on the magnitude of the Q value, the same path will be repeatedly followed and convergence will not occur, so it is desirable to use the ε greedy method, which makes random choices with a certain probability, when selecting an action. The ε greedy method does not simply make decisions based solely on the magnitude of the Q value, but has an element of random decision making with a certain rate, so it can act to shorten the total processing time even in cases where the reward is 0, not just in the second trial.

In addition to formula (1), the Q value may be calculated using the following formula (2) that takes into account the learning coefficient and discount rate. The following formula has the effect of preventing the Q value from diverging, and allows machine learning to proceed appropriately.
Q(s, a)←Q(s, a)+α(r+γmaxQ( _snext , _anext )
−Q(s, a)) ... (2)
Here, r: reward, s: state, a: action selected in state s, α: learning coefficient, γ: discount rate, and maxQ(s _next , a _next ): the maximum value of the Q value that can be taken in the next state.

The production planning system of the present invention may also simultaneously perform deep learning using neural networks in addition to machine learning.

As described above, according to this embodiment, by dividing the time that equipment is down into pre-setup time, work wait time, and post-setup time, it is possible to obtain a standard processing time with pre-setup time, work time, and post-setup time as one unit. Therefore, it is possible to provide a production planning system that accurately grasps the operating status and creates accurate production plans in a cell production method in which humans are involved and in a parallel process in which multiple pieces of equipment are operated.

REFERENCE SIGNS LIST 1 Production planning system 2 Computer 3 Replacement detection sensor 4 Work detection sensor 5 Touch panel 11, 12, 13 Operation status analysis section 21 Operation detection sensor 22 Indicator light 41 Processing instruction sheet 42 Monitor 51 Intermediate product 61 Tray 100 Worker _M1 , _M2 , _M3 Equipment

Claims (12)

A production planning system for creating a production plan for a process in which a plurality of pieces of equipment operated by workers process a plurality of intermediate products one unit at a time in a cell production system in which workers are involved or a parallel process in which a plurality of pieces of equipment are operated, comprising:
an input unit for inputting processing information and processing quantity for each type of the intermediate product;
an operating status analysis unit including: an operation detection sensor provided in each of the plurality of pieces of equipment, which determines a stop time and an operating time of the plurality of pieces of equipment; a replacement detection sensor which detects a replacement of the intermediate products in the plurality of pieces of equipment; and an operation detection sensor which detects whether the worker is working or waiting for work in the plurality of pieces of equipment;
a calculation unit which divides the downtime into pre-setup time, post-setup time, and work waiting time based on a signal from the operation status analysis unit when the processing of the same intermediate product is repeated, obtains a standard processing time indicating the processing time of the intermediate product for each type of product by using the pre-setup time, the operation time, and the post-setup time as one unit, and adjusts between the plurality of facilities to allocate the processing quantity;
an output unit that issues processing instructions to each piece of equipment based on the allocation result of the calculation unit;
A production planning system comprising: The production planning system according to claim 1, further comprising: a calculation unit that, when the type of the intermediate product being processed is changed, extracts a type changeover time from the stop time to obtain a standard type changeover time, and adjusts and allocates the processing quantity between the multiple pieces of equipment based on the standard processing time and the standard type changeover time. The production planning system according to claim 1 or 2, characterized in that the operation detection sensor is a sensor that detects the illumination of an indicator light provided on the equipment. The production planning system according to any one of claims 1 to 3, characterized in that the work detection sensor is a human presence sensor that is provided in front of the equipment and determines whether the worker is working. The production planning system according to any one of claims 1 to 3, characterized in that the work detection sensor is a safety device sensor provided in the equipment and determines whether the worker is working. The production planning system according to any one of claims 1 to 5, characterized in that the replacement detection sensor is a reader that reads the replacement timing of identification data sent to the process together with intermediate products or processing data sent to the equipment. The production planning system according to any one of claims 1 to 6, characterized in that the operation status analysis unit has a worker identification unit that identifies a worker, and the calculation unit calculates a standard processing time for each worker. The production planning system according to any one of claims 1 to 7, characterized in that the production planning system repeatedly reallocates the remaining intermediate products to be processed based on production results. The production planning system according to any one of claims 1 to 8, characterized in that the calculation unit has a simulator that predicts production quantities based on information from the input unit and the operation status analysis unit, and the calculation unit allocates intermediate products to be processed based on the results of the simulator. The production planning system according to claim 9, characterized in that the calculation unit has a learning unit that performs reinforcement learning. The production planning system according to claim 10, characterized in that the calculation unit performs reinforcement learning using Q-learning in which the reward is the time required to complete the processing quantity. The production planning system according to any one of claims 1 to 11, characterized in that the equipment includes one or more devices selected from the group consisting of processing machines, robots, PLCs, conveying machines, measuring instruments, testing equipment, presses, presses, printing machines, die-casting machines, injection molding machines, food machinery, packaging machines, welding machines, cleaning machines, painting machines, assembly machines, mounting machines, woodworking machines, sealing devices, and cutting machines.

Images