WO2021019702A1

WO2021019702A1 - Automated warehouse optimization system

Info

Publication number: WO2021019702A1
Application number: PCT/JP2019/029917
Authority: WO
Inventors: 基可高木; 尚久山田; 紀郎横山; 卓治渡部
Original assignee: トーヨーカネツ株式会社
Priority date: 2019-07-30
Filing date: 2019-07-30
Publication date: 2021-02-04
Also published as: JP6677858B1; JPWO2021019702A1

Abstract

The present invention comprises: a model generation means that generates, through machine learning, a learned model for defining a logistics process scenario for optimizing specific pieces of logistics process scale information from among logistics process score values, there being used, as teaching data, a past logistics process scenario that chronologically defines how all devices constituting a logistics system should operate in order to optimize, for logistics orders received in a given period, specific logistics process scale information accompanying processing of the logistics orders by the logistics system, and there also being used, as the teaching data, referenced logistics process scale information obtained by combining at least one from among throughput information, process time information, power consumption information, delivered-order number information, information pertaining to the number of personnel required for a process, and process number information per unit time for a special article to be processed that includes a cell component, these pieces of information being obtained as a result of operation according to the past logistics process scenario; a logistics order process reception means that receives all logistics orders in the given period; a past logistics process scenario acquisition means that acquires the past logistics process scenario that defined chronological behavior pertaining to all of the devices with respect to the logistics orders received by the logistics order process reception means; a scale information specification means by which desired scale information is specified from among the logistics process scale information; and a processing means that, on the basis of the past logistics process scenario acquired by the past logistics process scenario acquisition means and scale information specified by the scale information specification means, outputs a logistics process scenario using the learned model generated by the model generation means, the logistics process scenario chronologically defining how all of the devices constituting the logistics system should operate in order to optimize the scale information.

Description

Automated warehouse optimization system

The present invention relates to, for example, an automated warehouse optimization system in a three-dimensional automated warehouse installed in a distribution center, a learned model for that purpose, a distribution processing scenario estimation system, a distribution processing scenario output system, and a distribution warehouse control system.

In recent years, in the field of logistics, the desire to obtain the necessary products when needed has become stronger not only in industry but also among general consumers, so rapid rationalization of logistics is being promoted.

Among these, the warehouse management system (warehouse management system) is used in distribution centers, etc., which are responsible for consolidating products in a predetermined location in the warehouse for primary storage, leaving the storage location according to the order of the shipping destination, and delivering the products to the shipping destination. Warehouse
Attention is being paid to three-dimensional automated warehouses that simultaneously perform a series of product sorting operations (warehousing / delivery operations, storage operations, etc.) based on the Management System (WMS) (Patent Documents 1 to 4).

Japanese Patent No. 6185619 Japanese Patent No. 6231168 International Publication No. WO2013 / 46379 Pamphlet Japanese Patent No. 5508259

Generally, a three-dimensional automated warehouse is equipped with a warehouse means for storing a case in which goods are stored and a picking station for picking a required quantity of goods from the case discharged from the warehouse means and sending them to a shipping line.

The warehousing means is composed of a large number of shelves for storing cases, transportation means for transporting cases and storing them on the shelves, and transport means for warehousing from the shelves. Further, downstream of the picking station, a packing line for packing the picked products by an automatic sealing device or the like may be provided.

Conventionally, in this type of three-dimensional automated warehouse, past warehousing stored in a warehouse management system (WMS) is performed in order to efficiently perform a series of warehousing operations such as product arrival → warehousing / storage → warehousing → picking → packing / shipping. Simulates warehousing / delivery movements based on warehousing operation logs and attribute data (quantity, date, warehousing / shipping destination, storage location, weight, etc.) of product items, and uses this simulation model to optimize shipping efficiency. Work is being done to generate the conversion model.

However, in recent years, with the diversification and complexity of attribute data of product items, location management of the warehouse management system (WMS) has become more complicated, so an optimization model for improving delivery efficiency has been adopted. The amount of calculation required to generate it becomes excessive, and it becomes difficult to generate an optimized model using a computer that runs on a program (rule base) created based on rules that humans have thought of.

For example, looking at the work of delivering goods stored on the shelves of warehouse means by means of transportation, when optimizing the access route to the shelves and the order of picking goods according to the optimization model based on the rules, it is not always the case of transportation. The behavior of the means is not always optimal. In addition, in order to improve the optimization model based on the rule base in the three-dimensional automated warehouse in operation, large-scale work such as stopping the operation of the device and removing the stored products from the shelves is required.

The present invention has been made in view of such a problem of the prior art, and in a three-dimensional automatic warehouse managed by a warehouse management system (WMS), a request of a warehouse user (user) (for example, information on the number of waiting items for delivery). , Conveyor movement distance information, inter-shelf movement waiting number information, inter-shelf distance information, stacker crane moving distance information, relocation occurrence number information, warehousing waiting number information, passage distance information, buffer section waiting number information, allocationable number information An automated warehouse optimization system that optimizes at least one of these information or information obtained by combining two or more pieces of information at a predetermined ratio, a trained model for that purpose, a distribution processing scenario estimation system, and distribution processing. The purpose is to provide a scenario output system and a distribution warehouse control system.

In order to solve the above-mentioned problems, in the trained model according to the typical aspect of the present invention, for the physical distribution order received in a certain time, the specific physical distribution order is processed by the physical distribution system. Logistics processing scale To optimize information, make the computer function to output information that defines a physical distribution processing scenario that defines how all the devices that make up the physical distribution system should operate over time. This is a trained model for joining the input layer with the scenario as the input value from the past distribution processing scenarios specified for the distribution orders received in a certain period of time, and the input layer with a weighting coefficient. The input layer is provided with one or more intermediate layers and an output layer joined to the intermediate layer with a weighting coefficient, an operation based on the weighting coefficient is performed on the input layer, and the specific distribution process is performed from the output layer. Make the computer function to output information that defines the logistics processing scenario for optimizing the scale information.

Similarly, in order to solve the above-mentioned problems, in the physical distribution processing scenario estimation system according to the typical aspect of the present invention, the physical distribution order is processed by the physical distribution system for the physical distribution order received in a certain time. A past logistics processing scenario that defines how all the devices that make up the logistics system should be operated over time in order to optimize specific logistics processing scale information associated with the above, and the past logistics. Throughput information, processing time information, power consumption information, delivery completion order number information, number of personnel required for processing, processing number information per unit time of special processing target goods including sale items, obtained as a result of operation by the processing scenario Logistics processing scenario for optimizing a specific physical distribution processing scale information among the said physical distribution processing performance values by using as teacher data in light of the physical distribution processing scale information that is a combination of at least one or two or more of the above. A model generation means for generating a trained model that defines the above, a distribution order processing reception means for accepting all distribution orders in a certain time, and all the distribution orders received by the distribution order processing reception means. The past distribution processing scenario acquisition means for acquiring the past distribution processing scenario that defines the behavior of the device over time, the scale information specifying means for specifying the desired scale information from the distribution processing scale information, and the model. Using the trained model generated by the generation means, the scale information is optimized from the past distribution processing scenario acquired by the past distribution processing scenario acquisition means and the scale information specified by the scale information specifying means. Therefore, it is provided with a processing means for outputting a distribution processing scenario that defines how all the devices constituting the distribution system should be operated over time.

In the above embodiment, the distribution processing scale information includes throughput information, processing time information, power consumption information, delivery completion order number information, personnel number information required for processing, and special processing target article including sale items per unit time. It may be obtained in light of at least one or a combination of two or more of the processing number information, and the distribution processing scenario may include delivery waiting number information, conveyor movement distance information, and inter-shelf movement waiting. Based on at least one of number information, inter-shelf distance information, stacker crane movement distance information, relocation occurrence number information, warehousing waiting number information, passage distance information, buffer section waiting number information, and allocatable number information. It may be created.

Furthermore, in order to solve the above problems, in the physical distribution processing scenario output system according to a typical aspect of the present invention, it relates to machine learning as a scenario for processing physical distribution orders received in a certain time. The first physical distribution processing scenario output by the physical distribution processing scenario estimation system, the second physical distribution processing scenario based on the rule base which is a predetermined program, the first physical distribution processing scenario, and the second physical distribution. An estimation model that estimates which of the first or second distribution processing scenarios is more preferable for a specific distribution order by using the distribution processing scale information defined for each processing scenario of the processing scenario as teacher data. A model generation means for generating a model by machine learning, an input means for inputting a specific distribution order, and a distribution order information specifying means for specifying the distribution order regulation information that defines the specific distribution order input by the input means. The physical distribution specified by the physical distribution order information specifying means using the scale information specifying means for specifying the desired scale information from the physical distribution processing scale information and the estimation model generated by the model generating means. In order to optimize the scale information from the order regulation information, a processing means for outputting which of the first and second distribution processing scenarios is more preferable is provided.

Alternatively, as an alternative, the first means for obtaining the first physical distribution processing scenario output by the physical distribution processing scenario estimation system related to machine learning as a scenario for processing the physical distribution order received in a certain time, and the above-mentioned Based on a rule base, a scale information specifying means for specifying desired scale information from the physical distribution processing scale information and a second physical distribution processing scenario for optimizing the scale information specified by the scale information specifying means are used. In order to optimize the scale information for the second means obtained, the input means for inputting a specific distribution order, and the distribution order input by the input means, the first or second distribution processing scenario A determination means for determining which of the above is more preferable is provided.

Further, in order to solve the above problems, in the distribution warehouse control system according to the typical aspect of the present invention, distribution processing is performed based on the past distribution processing scenario defined for the distribution order received in a certain period of time. A physical distribution processing scenario that outputs an optimum physical distribution processing scenario for a specific physical distribution order using a model generation means that generates a trained model for defining a scenario by machine learning and a trained model generated by the model generation means. It includes an output means and a control unit that controls and / or drives a distribution warehouse by using the optimum distribution processing scenario output by the distribution processing scenario output means.

In the above, the “logistics processing scenario” is information that defines how all material handling equipment related to the physical distribution system is operated over time, and includes a warehousing log and / or a warehousing log. More specifically, as a component of this distribution processing scenario or an input for forming the distribution processing scenario, the number of items waiting to be delivered, the information on the conveyor movement distance, the information on the number of items waiting to move between shelves, and the information on the distance between shelves. , Stacker crane movement distance information, relocation occurrence number information, warehousing waiting number information, passage distance information, buffer section waiting number information, allocationable number information, etc. are included.
Further, the "logistics processing scale information" selects what is preferable when, for example, an AI is used to acquire a trained model, or when a scenario related to physical distribution processing is obtained using the acquired learned model. -Information related to the judgment criteria required to specify.

Also, "throughput" means the processing quantity per unit time.

Alternatively, the following aspects can be taken.

That is, in the automated warehouse optimization system according to the first aspect of the present invention, a single or a plurality of product items provided with the first identifier are stored, and a second identifier associated with the first identifier is provided. A storage means, a single or multiple shelves having one or more stages and for storing the storage means, a single or multiple passages arranged adjacent to the shelves, and the passages provided. , A warehouse means having a single or a plurality of transport means for transporting the storage means and storing them on the shelf and also warehousing from the shelf, and a required quantity from the storage means delivered from the warehouse means. A picking means for picking the product item and putting it into a shipping medium provided with a third identifier associated with the first identifier is arranged, and the storage means for which the picking of the product item is completed is provided. The warehouse means includes a picking station for returning to the warehouse means, and the warehouse means controls for associating the first identifier provided for the product item with the second identifier provided for the storage means, and the storage. Control for storing the means for storage on the shelves and control for issuing a warehousing command for unloading the means for warehousing from the warehousing means are performed, and at the picking station, the required quantity of the means is said to be warehousing. An input completion command based on the control of a picking command for picking a product item and the association between the first identifier provided on the picked product item and the third identifier provided on the shipping medium. The control and the control of the return command for returning the storage means for which the picking of the product item is completed to the warehouse means are performed.

In the first aspect, the automated warehouse optimization system according to the second aspect of the present invention further includes a packing means for taking out the product item put into the shipping medium and packing it with a packing material.

In the automated warehouse optimization system according to the third aspect of the present invention, in the first or second aspect, the attribute data of the product item handled by the warehouse means and the picking station is obtained by the warehouse management system (WMS). Be managed.

In the automated warehouse optimization system according to the fourth aspect of the present invention, in any one of the first to third aspects, the transport means simultaneously combines horizontal and / or vertical movement of the passage. It is a stacker crane type transport means that moves vertically.

In the automated warehouse optimization system according to the fifth aspect of the present invention, in any one of the first to fourth aspects, the warehouse means includes an elevating means for vertically moving the storage means and the shelf. Further, a buffer means for transferring the storage means between the transport means and the elevating means is provided at the end of each stage of the above, and the buffer means includes the transport means and the elevating means. Control is performed to prevent mutual interference between the storage means passed between them.

In the automated warehouse optimization system according to the sixth aspect of the present invention, in any one of the first to fifth aspects, the warehouse means stores a plurality of types of highly correlated product items in advance in the same manner. Control (co-occurrence article aggregation) for storing in the means is performed.

The automated warehouse optimization system according to the seventh aspect of the present invention comprises, in any one of the first to sixth aspects, the plurality of storage means for storing a single or a plurality of highly correlated product items. Control is performed for warehousing in the same shelf.

The automated warehouse optimization system according to the eighth aspect of the present invention has at least one of the control performed by the warehouse means and the control performed by the picking station in any one of the first to seventh aspects. One or more is based on at least one optimization model, including an optimization model generated by artificial intelligence (AI) technology.

In the eighth aspect, the automated warehouse optimization system according to the ninth aspect of the present invention artificially performs at least one of the control performed by the warehouse means and the control performed by the picking station. When based on an optimization model generated by artificial intelligence (AI) technology, the evaluation and verification of the control is performed in a simulated environment.

The automated warehouse optimization system according to the tenth aspect of the present invention comprises the optimization model generated by the artificial intelligence (AI) technique and the non-artificial intelligence (non-AI) technique in the eighth or ninth aspect. The optimization model generated by the second artificial intelligence (AI) technology different from the artificial intelligence (AI) technology or the second non-artificial intelligence (non-AI) different from the non-artificial intelligence (non-AI) technology. Includes steps to compare and evaluate by technology.

The automated warehouse optimization system according to the eleventh aspect of the present invention has the artificial intelligence (AI) technique in any of the eighth to tenth aspects prior to the step of generating an optimization model by the artificial intelligence (AI) technique. Determine whether the (AI) technology should generate the optimization model or the non-artificial intelligence (non-AI) technology should generate the optimization model, and the non-artificial intelligence (non-AI) technology should generate the optimization model. If determined, at least one of the controls performed by the warehouse means and the controls performed by the picking station is based on an optimization model generated by non-artificial intelligence (non-AI) technology. Is done.

In the eleventh aspect of the automated warehouse optimization system according to the twelfth aspect of the present invention, the determination is made by artificial intelligence (AI) technology or non-artificial intelligence (non-AI) technology.

In the automated warehouse optimization system according to the thirteenth aspect of the present invention, a single unit product item provided with a first identifier is stored, and a second identifier associated with the first identifier is provided for storage. An individualizing means, a single or a plurality of shelves having one or a plurality of stages and storing the individualizing means for storage, a single or a plurality of passages arranged adjacent to the shelves, and the above. A warehouse means provided in a passage and having one or a plurality of transport means for transporting the individualized means for storage and storing them on the shelf and also leaving the shelves, and the storage delivered from the warehouse means. A picking means for picking a product item of the single unit from the individualization means and putting it into a shipping medium provided with a third identifier associated with the first identifier is arranged, and the single unit is arranged. The warehouse means includes a picking station for sending the storage individualizing means for which picking of the product item has been completed to peripheral equipment, and the warehouse means has the first identifier provided for the product item of the single unit and the storage. A control for associating the second identifier provided in the individualizing means for storage, a control for storing the individualizing means for storage on the shelf, and the warehouse means for storing the individualizing means for storage. The picking station controls a picking command for picking a product item of the single unit from the storage individualizing means, and controls the picking command for picking the goods from the warehouse. The control of the input completion command based on the association between the first identifier provided on the product item of one unit and the third identifier provided on the shipping medium, and the picking of the single unit product item are completed. Control of the return command for sending the storage means to the peripheral equipment is performed.

In the thirteenth aspect of the automated warehouse optimization system according to the fourteenth aspect of the present invention, at least one or more of the control performed by the warehouse means and the control performed by the picking station is artificial. It is based on at least one optimization model, including an optimization model generated by artificial intelligence (AI) technology.

According to the automated warehouse optimization system of the present invention, it is possible to improve the delivery efficiency of goods in a three-dimensional automated warehouse managed by a warehouse management system (WMS).

It is a conceptual diagram of an example of the three-dimensional automated warehouse which concerns on one Embodiment of this invention. It is a perspective view which shows a part of the storage warehouse part of the three-dimensional automated warehouse shown in FIG. It is a block diagram which shows the system configuration of WMS and WCS which manages a three-dimensional automated warehouse shown in FIG. It is a perspective view of an example of the storage case which is stored in the storage warehouse part of the three-dimensional automated warehouse shown in FIG. It is a tray table which shows the association information of the 1st identifier provided in the product item and the 2nd identifier provided in the storage case. It is a warehousing table showing information associating the first identifier provided for the product item with the arrival source. It is a delivery table showing information associating the first identifier provided for the product item with the delivery destination. It is a storage data table that records information on product items. It is a conceptual diagram of an example of the optimization model generation step which concerns on one Embodiment of this invention. It is a conceptual diagram of an example of the optimization model generation step which concerns on one Embodiment of this invention. It is a flowchart which shows an example of the operation which concerns on FIG. It is a flowchart which shows an example of the operation which concerns on FIG.

Hereinafter, the automated warehouse optimization system according to the embodiment of the present invention will be described with reference to the drawings. In the following, the range necessary for the explanation for achieving the object of the present invention will be schematically shown, and the range necessary for the explanation of the relevant part of the present invention will be mainly described. It shall be based on known technology.

FIG. 1 is a conceptual diagram of an example of a three-dimensional automated warehouse according to the present embodiment, FIG. 2 is a perspective view showing a part (one row of shelves) of a storage warehouse section of the three-dimensional automated warehouse shown in FIG. , Is a block diagram showing a system configuration of WMS and WCS that manages the three-dimensional automated warehouse shown in FIG.

The three-dimensional automated warehouse 1 includes a storage warehouse section (warehouse section) 100, a picking station 200, and a packing section 300.

The storage warehouse unit 100 includes a plurality of rows of shelves 10 in which a plurality of stages having storage spaces are arranged along the aisles, and a plurality of passages 20 arranged adjacent to each shelf 10 (the aisles 20 include shelves 10). (Including a mode in which the shelves are sandwiched between the shelves 10) and a transport unit 30 provided in each aisle 20 and moving along the aisle 20.

Further, in the storage warehouse section 100, an elevating section 50 which is arranged at the end of each passage 20 and moves the storage case 40 in the vertical direction, and a transport section 30 which is arranged at the end of each stage of the shelf 10. A buffer conveyor (buffer unit) 60 for delivering and receiving the storage case 40 to and from the elevating unit 50 is further provided.

The transport unit 30 is, for example, a stacker crane type transport device that simultaneously moves the passage 20 in the horizontal direction and / or the vertical direction, and is a storage case (storage unit) in which a plurality of product items 70 are stored. The 40 is supported by a stretchable arm or the like to be stored in the shelf 10 or discharged from the shelf 10.

The attribute data (quantity, date, receipt / shipment destination, storage location, weight, etc.) of all the product items 70 to be sorted in the three-dimensional automatic warehouse 1 is the host computer (or server, cloud, etc.) that collectively manages the warehouse operations. It is managed by a WMS (warehouse management system) 400 equipped with the above and a WCS (warehouse control system) 500 incorporating a computer connected to the host computer of the WMS 400.

As shown in FIG. 3, the WCS500 includes a CPU, main memory, and an external interface, and has a computer connected to the host computer of the WMS400 and a storage connected to the computer via the external interface. Remote operation, communication line, etc. by wireless and / or wired to the material handling equipment (conveying unit 30, elevating unit 50, buffer conveyor 60, etc.), picking station 200, packing unit 300, etc. in the three-dimensional automatic warehouse 1 according to the instruction from Work instructions such as warehousing, transportation, warehousing, picking, and automatic boxing are issued through.

As shown in FIG. 4, a single or a plurality of product items 70 provided with a first identifier 71 made of a barcode are stored inside the storage case 40, and the outer portion of the storage case 40 contains a plurality of product items 70. , A second identifier 41 consisting of a barcode that can be associated with the first identifier 71 of the product item 70 is provided.

The first identifier 71 is for identifying the product item 70, and is different for each product. The second identifier 41 is for identifying the storage case 40, and is different for each storage case 40. The first identifier 71 and the second identifier 41 may be identifiers other than barcodes, such as a two-dimensional code and RFID.

Although not shown, the distribution center equipped with the three-dimensional automated warehouse 1 is equipped with a truck berth where product items 70 from each manufacturer arrive. When the product item 70 from each manufacturer arrives at the truck berth of the distribution center, a tray-making operation is performed to unpack each product item 70 from a packaging container such as a cardboard box and store it in the storage warehouse section 100.

The tray-making work is a process of associating the first identifier 71 attached to each product item 70 with the second identifier 41 attached to the storage case 40. This is the association described above. When the product item 70 is a stationery item, it is divided into one that is stored in the storage case 40 one by one like a stapler and one that is stored in a box unit like a pencil.

The second identifier 41 attached to the storage case 40 is read when the storage case 40 passes in front of the barcode reader provided in the tray-making work place. For reading the second identifier 41, a known technique such as irradiating an LED light source and receiving the reflected light with a photodiode is used.

On the other hand, the first identifier 71 attached to the product item 70 is read by a handy scanner or the like provided at the product loading location of the truck berth every time the product item 70 is put into the storage case 40. Then, the computer of WCS500 creates a tray table shown in FIG. 5 based on the read first identifier 71, its quantity, and the second identifier 41, and stores the tray table in the memory area for the tray table in the storage.

As a result, the names and numbers of the product items 70 stored in each storage case 40 are grasped by the WCS500. In the tray making work, in addition to the case where the product item 70 is put into the empty storage case 40, if there is a product putting space in the storage case 40 in which the same product item 70 has already been put, the storage is performed. After the use case 40 is stored in the shelf 10, it may be transported to the tray-making work place.

Further, the WCS500 has a second identifier 41, a type and quantity of product items 70, a number, a stage number, a warehousing time, etc. of the shelf 10 of the storage warehouse unit 100, which are required when performing automatic warehousing / delivery work in the storage warehouse unit 100. Create a storage data table as shown in FIG. 6 in which the information of the above is recorded. This storage data table is stored in the storage data table memory area in the storage of the WCS500, and the information is updated as necessary at the time of delivery, picking work, packing completion, and the like. These information (including log information) recorded in the stored data table can be used as basic data for machine learning by artificial intelligence (AI) described later.

For example, when the storage case 40 is stored in the storage warehouse unit 100, the CPU of the computer of the WCS500 is based on the address information including the shelf number, the stage number, the passage number, the storage time, etc. of the stored storage case 40. The stored data table is updated, and the updated stored data table is stored in the stored data table memory area in the storage.

Further, when the storage case 40 stored in the storage warehouse unit 100 is returned to the tray-making work place and an additional product item 70 is added, or when the product item in the storage case 40 delivered from the storage warehouse unit 100 is added. When the storage case 40 is returned to the storage warehouse unit 100 after a part of the 70 is picked at the picking station 200, the information in the traying table is also updated by the CPU of the computer of the WCS500.

The storage case 40 for which the tray-making work described above has been completed is transported to the storage warehouse section 100 by the storage conveyor 600 (see FIG. 2). Since one end of the warehousing conveyor 600 that connects the tray-making work place and the storage warehouse unit 100 is arranged above the picking station 200, in FIG. 1, in order to make the configuration of the picking station 200 easier to see, warehousing The illustration of the conveyor 600 is omitted.

When the storage case 40 is stored in the storage warehouse unit 100, first, the storage command issued by the host computer of the WMS 400 is transmitted to the computer of the WCS 500. The warehousing order includes a warehousing table associated with the name, quantity, warehousing time, etc. of the product item 70.

The CPU of the WCS500 grasps the name and quantity of the product item 70 to be stored in the storage warehouse unit 100 by the above storage order. The CPU of the WCS500 compares the storage data table and the warehousing table based on a predetermined work instruction program stored in the hard disk drive, creates the optimum warehousing order information, and creates the material handling equipment (conveyor unit) of the storage warehouse unit 100. 30, buffer conveyor 60, elevating part 50, etc.) are given work instructions. Here, the warehousing order information is a route and an order for determining in which order and in which frontage each storage case 40 storage case 40 is stored.

At this time, the CPU of the WCS500 refers to the warehousing table and the traying table, confirms which storage case 40 each product item 70 is stored in, and then compares the storage information table and the traying table. Then, it is determined in which stage of which shelf 10 each storage case 40 is stored.

When determining the address for storing the storage case 40 (the stage of the shelf 10), for example, the address information of the storage case 40 that has already been stored is read from the storage data table, and the shelf 10 with a small number of storage cases 40 is read. Make sure to store in the stage. At that time, it is preferable to store the items in order from the shelf 10 stage closest to the storage conveyor 600.

Each storage case 40 for which a warehousing order has been issued in this way moves to the stage of the shelf 10 determined in order according to the operation of the material handling device and is warehousing. Then, the WCS500 updates the storage data table according to the storage data (address information of the storage case 40, storage time, etc.).

Here, as a criterion for determining which shelf 10 and which stage the storage case 40 is to be stored, for example, the number of storage cases 40 already stored in the shelf 10 has been focused on. Instead, attention may be paid to the occupancy rate of the storage case 40 with respect to the shelf 10 per unit time.

The automated warehouse optimization system of the present invention includes control in the tray-making process described above, control of material handling equipment (conveyor unit 30, buffer conveyor 60, elevating unit 50, etc.) in the storage warehouse unit 100, picking process and / or packing described later. As a method for optimizing the control in the process, the optimization model generated by artificial intelligence (AI) is compared with the optimization model generated by the rule base (a program created based on the rules thought by humans). Adopt a method to use both properly as needed.

The rule-based program is stored in the storage of WCS500. The artificial intelligence (AI) program is stored in a hard disk drive or the like of a server which may be installed in a place different from the WCS500, and is connected to the computer of the WCS500 via a communication unit. The artificial intelligence (AI) program may be stored in the WCS500 storage independently of the rule-based program.

In addition, each of the rule base and artificial intelligence (AI) operates with a correct judgment as a material handling device in the storage warehouse unit 100, an operation device for operating various devices in the picking station 200 and the packing unit 300, and the operation. It is composed of a judgment device that makes a judgment as to whether or not to do so. The determination device has predetermined priorities such as shortest time delivery, shortest distance delivery, minimum power, priority delivery information for preferentially issuing a specific product that may be predetermined, and the number of product orders to be delivered. Optimal solution based on priority items selected from picking staffing, throughput (processing amount per unit time), sale (selling information and campaigns specific to days of the week, seasons, etc.), degree of response to requests, trends, etc. It is determined whether the output is a rule-based operating device or an artificial intelligence (AI) operating device.

Here, artificial intelligence (AI) is based on iterative learning of data about a specific event, finding and modeling features (regularity and relationships) from the results, and based on the optimum solution obtained from this model. It means all methods that use learning algorithms to make judgments and predictions, not only machine learning that analyzes data input to a computer based on algorithms such as supervised learning, unsupervised learning, and deep learning, but also machines. It is a concept that includes deep learning, which is an extension of learning.

Deep learning is a method of learning data using a neural network modeled on human nerve cells. A neural network is a computer model of nerve cells in the human brain. It is an input layer (input), a hidden layer (also called an intermediate layer, which can be multi-layered), and an output layer (output). Consists of. That is, deep learning is defined as machine learning using a multi-layered neural network in which many hidden layers exist. There are many types such as convolutional neural networks, recurrent neural networks, and long / short-term storage unit neural networks.

Other artificial intelligence (AI) methods include, but are not limited to, the K-means method, the Bayesian network method, the Kalman filter method, the support vector machine method, the decision tree method, and the like.

FIG. 7 is a conceptual diagram showing an example of the optimization model generation method according to the present embodiment.

Here, it is recorded in the past warehousing / delivery operation log stored in the storage of WCS500, the operation log of the material handling device, the attribute data of the product item (quantity, date, receipt / shipment destination, storage location, weight, etc.), and the storage data table. The shortest movement time (time saving model) of the material handling equipment (conveying unit 30, buffer conveyor 60, elevating unit 50, etc.) is generated by the rule base created based on the above data, etc., and input is performed based on the above data, etc. An example is shown in which a variable (explanatory variable) is extracted and the shortest travel time (time reduction model) of a material handling device is generated by artificial intelligence (AI) technology such as deep learning.

When the above control performed by the storage warehouse unit 100 and the picking station 200 is optimized by artificial intelligence (AI), each control may be optimized individually, or the entire control may be optimized collectively. Further, as to which control is prioritized and optimized, a person may give a weight to each control according to the situation, or the system may give a weight automatically converted from a parameter describing the situation. ..

If the system optimizes each control individually, co-occurrence article aggregation (for example, if you order a pencil, you will also order an eraser (if the rate is above a certain level, you will find pairing and specify the pairing rules). In the case of (including the fact that the items to be paired are put in the same tray), for example, mutual information or unsupervised machine learning using Co-occurrence's self-organizing map (SOM) can be considered. Here, SOM means, for example, to generate a cluster related to co-occurrence articles, and is described in detail in the following document, for example (Kohonen T. Self-organizing formation of topologically correct).
feature maps. Biol. Cybern., 43, 1982, 59-69.). Further, in the case of control for storing the storage case 40 in the shelf 10 and leaving the shelf 10, for example, the processing amount per unit time of the transport unit 30, the elevating unit 50, and the buffer conveyor 60 is regarded as a route search problem. It is conceivable to perform the calculation of the shortest path search by a simple combination logic method or the traveling salesman problem method using a Hopfield network or the like.

When optimizing the entire control, it is regarded as an optimum point search problem in the n-dimensional space composed of control parameters, and this is treated as a curve approximation problem in the n-dimensional space to find the minimum value in the n-dimensional space. You may go.

In addition, regarding the optimization method according to the situation, the importance of the parameters that describe the situation can be set by a human or the system, and the importance of the interdependent optimization parameters can be determined. Simply, if the parameter value 1 of interest optimizes the optimization control item A but not the optimization control item B, while the parameter value 2 of interest optimizes the optimization control item A. However, it is assumed that there is a problem as to which optimization control item should be adopted, such as optimizing the optimization control item B. In this case, if the importance of the optimization control item A is high, a method of completely ignoring the optimization of the optimization control item B and adopting the parameter value 1 or considering each parameter as an n-dimensional input and energy. It is conceivable to use a machine learning method that converges to the minimum point, or a supervised machine learning method that shows desirable results for humans.

When generating the shortest travel time (time saving model) of the material handling equipment in the storage warehouse unit 100, the artificial intelligence (AI) and the rule base are the number of storage cases 40 stored in each shelf 10 and the storage for each shelf 10. Considering the occupancy rate of the case 40, reading the combination of the product items 70 that can be sold together from the past shipping history, and arranging the plurality of types of product items 70 that are considered to be sold together on the same shelf 10. At least one of them or a combination thereof may be considered.

For example, a plurality of types of product items 70, which are generally handled in stores of the same industry, such as a product group for bookstores, a product group for pharmacies, and a product group for electric stores, are aggregated into the same or close shelves 10 as product groups in the same category. Deploy. This is because it is easy to aggregate for the same GTP destination (it can be said that it is an example of co-occurrence goods aggregation), or in consideration of the best-selling products for each season, the best-selling product items 70 are distributed on a plurality of shelves 10. The product items 70, which are often sold together with the distributed hot-selling product items 70, are placed on the same shelf 10. The work load can be leveled.

Then, based on the actual delivery command per unit time such as one day, it is calculated how much of the product items 70 sold as a set are sold together, for example, 50% or more is artificial intelligence (AI). In the case of the product set predicted by, the warehousing order information derived by artificial intelligence (AI) can be adopted. On the other hand, if the number of sets different from the set predicted to sell together with artificial intelligence (AI) falls below 50%, it is possible to switch to adopt the warehousing order information derived by the rule base from the next day. ..

Then, while adopting the warehousing order information derived by the rule base, the prediction of the product set that can be sold together derived by artificial intelligence (AI) is continued, for example, the actual value per unit time such as one day. If 50% or more of the products sold as a set are the product set predicted by artificial intelligence (AI), the warehousing order information derived by artificial intelligence (AI) will be adopted from the next day. Switch to.

In addition, after warehousing the product item 70, the artificial intelligence (AI) predicts the warehousing, and when there are few warehousing orders and warehousing orders, the position of the storage case 40 is set so that the number of processing of the warehousing order per unit time during busy hours increases. The storage case 40 may be moved to a position where it can be easily delivered at the time of delivery.

For example, at the time of warehousing, a product group of the same category is used for each product handled in stores of the same industry, such as a product group for bookstores, a product group for pharmacies, and a product group for electric stores, and the products are grouped on the same or close shelves 10. Even if they are arranged, they may be arranged in a distributed manner during repeated warehousing and delivery. In that case, for example, a group of products dispersed at night when warehouse work is not performed may be aggregated on the same or close shelves 10.

When a customer places an order for the product item 70 stored on the shelf 10, the storage case 40 containing the product item 70 is taken out from the shelf 10 by the transport unit 30, and the buffer conveyor 60 and the elevating unit 50 are removed. It is transported to the picking station 200 via.

When the product item 70 is delivered from the storage warehouse unit 100, the WMS400 first issues a delivery order for the product item 70, which is transmitted to the picking station 200 via the computer of the WCS500. The delivery order includes the delivery information associated with the product item 70, its quantity, the delivery time, and the delivery destination.

The CPU of the WCS500 compares the storage information and the delivery information according to the default work instruction program stored in the storage, creates the optimum delivery order information, and creates the material handling equipment (conveyor unit 30, buffer conveyor 60, elevating unit 50, etc.). ) Is operated.

Specifically, in order to select the storage case 40 in which the product item 70 corresponding to the delivery instruction is stored according to the default work instruction program stored in the storage of the WCS500, and to take out the storage case 40 from the shelf 10. The transport unit 30 is moved to. Then, the transport unit 30 that has arrived at the storage case 40, which is the target of delivery, grips the storage case 40 and transfers it to the buffer conveyor 60.

At this time, the artificial intelligence (AI) is issued (for the day / morning / one hour) so that the number of processing orders per unit time can be maximized, for example, based on the issue history for the past week. Create a program or shipping order information. And if the same shipping trend continues in the past week, artificial intelligence (AI) will either improve the number of processing orders per unit time or maintain a high number of processing orders per unit time.

On the other hand, when the artificial intelligence (AI) has not learned about the sale in advance, such as when a new temporary sale (campaign) period is started on the day, the issue information in a state other than the sale period is output, so the unit time The number of orders per win may be extremely deteriorated. In that case, the issue information generated by the rule base is adopted instead of the issue information generated by artificial intelligence (AI).

In addition, if an irregular situation occurs, such as when the sale period that lasted for the past week has expired, or if any of the material handling equipment in the storage warehouse section 100 has failed, the delivery will be generated by artificial intelligence (AI). Since the issue efficiency generated by the rule base exceeds the efficiency, the rule-based issue information is adopted.

When the storage case 40, which is the target of the delivery order, is delivered from the storage warehouse unit 110, the WCS500 associates the address information consisting of the shelves 10 and the column numbers with the information of the storage case 40, and stores the storage case 40. The fact that the case 40 has been delivered is recorded in the storage data table together with the date and time, and the storage data table is updated.

When the storage case 40 delivered from the storage warehouse unit 110 arrives at the picking station 200, the picking unit 90 such as a worker or a robot displays a display screen displaying the product items 70 in the storage case 40 and their quantities. The product item 70 is picked according to the instructions of the above and put into an empty shipping case (shipping medium) 80 (or a packing material such as corrugated cardboard) flowing through the shipping conveyor 91.

When the loading of the product item 70 into the shipping case 80 is completed, the picking unit 90 touches the picking completion button displayed on the display screen. As a result, the WCS 500 transports the shipping case 80 toward the packing unit 300 according to the default work instruction program stored in the storage.

Although not shown, for example, the outer surface of the shipping case 80 is provided with a third identifier associated with the first identifier 71 of the product item 70 to be put into the shipping case 80. The third identifier includes information such as customer information and store information of the shipping destination, the name and name of the product item 70, and the date and time of delivery. In this case, not only the ID may be attached to the outer surface of the case, but the ID may be printed on the delivery note and included in the package, or the ID may be printed on the label and attached.

The WCS 500 associates these information described in the third identifier of the shipping case 80 with the information of the first identifier 71 of the product item 70 put into the shipping case 80, and the shipping case 80 is packed. The fact that the product has been transported to the unit 300 is recorded in the storage data table together with the date and time, and the storage data table is updated.

Of the storage cases 40 in which the product items 70 are picked at the picking station 200, the storage cases 40 in which the product items 70 remain inside are transported to the storage warehouse unit 100 again by the buffer conveyor 60 and the elevating unit 50. It is returned to the shelf 10 by the transport unit 30. In this case, the place where the storage case 40 is returned does not have to be the original place, and the vacant storage space of any shelf 10 is used (free location method).

On the other hand, the storage case 40, which has been emptied after the picking of the product item 70 is completed, is transported to the tray-making work place upstream of the storage warehouse unit 100, and is stored after the newly arrived product item 70 is stored. It is conveyed to the storage warehouse section 100 through the conveyor 600.

In this way, the picking station 200 controls the picking command for picking the required quantity of the product item 70 from the storage case 40 according to the instruction of the warehouse control system (WCS) 500, and the picked product item 70. Control of the input completion command based on the association between the first identifier 71 provided and the third identifier provided in the shipping case 80, and the storage case 40 in which the picking of the product item 70 is completed is sent to the storage warehouse unit 100. Control of the return command for returning, pairing control of picking units 90 having different skill levels, and the like are performed.

Artificial intelligence (AI) derives the optimum solution for most efficient control at the picking station 200 described above. Then, when this optimum solution is superior to the optimum solution derived by the rule base, the optimum solution derived by artificial intelligence (AI) is adopted.

In the packing unit 300, the product item 70 is taken out from the shipping case 80 sent from the picking station 200, packed in a packing material such as a cardboard box, and then subjected to shipping and delivery processing.

The shipping case 80 sent to the packing unit 300 includes shipping information read from the storage of the WCS 500, product items 70 and their quantities in the shipping case 80, customer information and store information of the shipping destination, and shipping. It is checked whether there is any discrepancy with the information such as the date and time, and then the product item 70 in the shipping case 80 is taken out and packed by an automatic sealing device or the like.

The WCS500 records the above-mentioned packing process information in the storage data table and updates the storage data table.

Artificial intelligence (AI) derives the optimum solution for the most efficient work in the packing section 300 described above. Then, when this optimum solution is superior to the optimum solution derived by the rule base, the optimum solution derived by artificial intelligence (AI) is adopted.

For example, when the product item 70 is packed by the automatic sealing device or the like, the packing unit 300 automatically cuts the packing base paper such as cardboard according to the size of the product item 70. At that time, the artificial intelligence (AI) determines whether to automatically cut the packing base paper according to the delivery order of the product item 70 or to automatically cut the packing base paper in units of the size of the packing base paper. As a result, it is possible to prevent waste of packing the product item 70 in a cardboard box larger than necessary and an error of inserting the product item 70 into a cardboard box smaller than the size of the product item 70.

In addition, the automated warehouse optimization system of the present invention compares an optimization model generated by artificial intelligence (AI) with an optimization model generated by a rule base (a program created based on a rule thought by humans). However, it includes a method to use both properly as needed.

For example, as shown in FIG. 8, when determining whether to obtain the shortest travel time of the Matehan device provided in the storage warehouse unit 100 on a rule basis or by artificial intelligence (AI), other than the control items performed by the storage warehouse unit 100. Two optimization models (artificial) by deep learning 2 (or non-artificial intelligence (non-AI) technology such as rule base) with the items (warehousing efficiency, worker allocation, environmental load, power consumption, etc.) as input variables The shortest travel time may be derived by a meta-optimization method that compares and evaluates an optimization model derived from intelligence (AI) and an optimization model derived from a rule base).

For the rule base, such meta-optimization may be performed by determining an evaluation function and adopting the parameters of each lower optimization method that gives the optimum solution. In addition, artificial intelligence (AI) may be performed by supervised machine learning or the like in which a human teaches whether or not the result is optimal.

In addition, prior to the step of generating a control optimization model with artificial intelligence (AI), should the optimization model be generated with artificial intelligence (AI) or optimized with non-artificial intelligence (non-AI) technology such as rule base? It is determined whether the model should be generated, and if it is determined that the optimized model should be generated by non-artificial intelligence (non-AI), the control of the Matehan equipment performed by the storage warehouse unit 100, the picking station 200 and the packing At least one or more of the various controls performed by the unit 300 may be performed based on an optimization model generated by a non-artificial intelligence (non-AI) technique. At that time, the determination of whether the optimization model should be generated by artificial intelligence (AI) or the optimization model by non-artificial intelligence (non-AI) is determined by artificial intelligence (AI) different from the above artificial intelligence (AI). ) Or non-artificial intelligence (non-AI).

According to the present embodiment as described above, the three-dimensional automated warehouse is provided by performing at least a part of the sorting work of the product item 70 performed in the three-dimensional automated warehouse according to the optimization model generated by artificial intelligence (AI). It is possible to improve the delivery efficiency of products at distribution centers and the like.

Therefore, according to the automated warehouse optimization system according to the present invention, it is possible to obtain an overall optimal solution in physical distribution as well as a physical distribution center equipped with a three-dimensional automated warehouse. In addition, the rationalization of the entire product distribution can be promoted.

In the above-described embodiment, the storage warehouse unit 100 for storing the product item 70 in the shelf 10 and discharging the product item 70 from the shelf 10 by using the stacker crane type transport unit 30, the buffer conveyor 60, and the elevating unit 50 will be described. However, the method of transporting the product item 70 in the storage warehouse unit 100 is not limited to this, and for example, the product item 70 can be stored in the shelf 10 using only the stacker crane type transport unit 30, or the shelf 10 You may leave the warehouse from. In addition, the product item 70 may be loaded and unloaded using a transport means other than the stacker crane method.

1: Three-dimensional automated warehouse 10: Shelf 20: Aisle 30: Transport section 31: Arm 40: Storage case (storage section)
41: Second identifier 50: Elevating part 60: Buffer conveyor (buffer part)
70: Product item 71: First identifier 80: Shipping case (shipping medium)
90: Picking section 91: Shipping conveyor 100: Storage warehouse section (warehouse section)
200: Picking station 300: Packing unit 400: Warehouse management system (WMS)
500: Warehouse control system (WCS)
600; Conveyor for warehousing

Claims

In order to optimize the specific distribution processing scale information associated with processing the distribution order by the distribution system for the distribution order received in a certain period of time, all the devices constituting the distribution system are used over time. A trained model for making a computer work to output information that defines a logistics processing scenario that defines how it should work.
An input layer that uses the scenario as an input value from the past distribution processing scenario specified for the distribution order received in a certain time, and
With one or more intermediate layers joined to the input layer with a weighting coefficient,
An output layer bonded to the intermediate layer with a weighting coefficient is provided.
Learning to make a computer function so as to perform an operation based on a weighting coefficient on the input layer and output information defining a physical distribution processing scenario for optimizing the specific physical distribution processing scale information from the output layer. Finished model.
In order to optimize the specific distribution processing scale information associated with processing the distribution order by the distribution system for the distribution order received in a certain period of time, how all the devices constituting the distribution system are processed over time. The past logistics processing scenario that stipulates whether or not to operate, and the throughput information, processing time information, power consumption information, delivery completion order number information, and personnel required for processing, which are obtained as a result of operating by the past logistics processing scenario. The physical distribution processing results are based on the physical distribution processing scale information in which at least one or two or more of the numerical information and the processing number information per unit time of the specially processed article including the sale item are combined as the teacher data. A model generation means that uses machine learning to generate a trained model that defines a physical distribution processing scenario for optimizing specific physical distribution processing scale information among the values.
Logistics order processing reception means that accepts all logistics orders in a certain time,
A past distribution processing scenario acquisition means for acquiring a past distribution processing scenario that defines the temporal behavior of all the devices for the distribution order received by the distribution order processing receiving means, and a past distribution processing scenario acquisition means.
Scale information specifying means for specifying desired scale information from the physical distribution processing scale information,
Using the trained model generated by the model generation means, the scale information is optimized from the past distribution processing scenario acquired by the past distribution processing scenario acquisition means and the scale information specified by the scale information specifying means. A physical distribution processing scenario estimation system including a processing means for outputting a physical distribution processing scenario that defines how all the devices constituting the physical distribution system should be operated over time.
The distribution processing scale information includes at least throughput information, processing time information, power consumption information, delivery completion order number information, personnel number information required for processing, and processing number information per unit time of special processing target articles including sale items. The logistics processing scenario estimation system according to claim 2, which is obtained in light of one or a combination of two or more of them.
The logistics processing scenario includes delivery waiting number information, conveyor movement distance information, inter-shelf movement waiting number information, shelf-to-shelf distance information, stacker crane moving distance information, rearrangement occurrence number information, warehousing waiting number information, passage distance information, and buffer. The physical distribution processing scenario estimation system according to claim 2 or 3, which is created based on at least one of the number of waiting parts information and the available number of allocation information.
Based on the first logistics processing scenario output by the logistics processing scenario estimation system related to machine learning as a scenario for processing the logistics orders received in a certain time, and the rule base which is a predetermined program. The second physical distribution processing scenario, the physical distribution processing scale information defined for each of the first physical distribution processing scenario and the processing scenario of the second physical distribution processing scenario are used as teacher data for a specific physical distribution order. A model generation means for generating an estimation model for estimating which of the first or second distribution processing scenarios is more preferable by machine learning, and
An input method for entering a specific distribution order,
A physical distribution order information specifying means that specifies the physical distribution order regulation information that defines the specific physical distribution order input by the input means, and
Scale information specifying means for specifying desired scale information from the physical distribution processing scale information,
In order to optimize the scale information from the distribution order regulation information specified by the distribution order information specifying means using the estimation model generated by the model generation means, the first or second distribution processing scenario A physical distribution processing scenario output system including a processing means for outputting which is more preferable.
As a scenario for processing a physical distribution order received in a certain time, a first means for obtaining a first physical distribution processing scenario output by a physical distribution processing scenario estimation system related to machine learning, and
Scale information specifying means for specifying desired scale information from the physical distribution processing scale information,
A second means for obtaining a second distribution processing scenario for optimizing the scale information specified by the scale information specifying means based on a rule base, and
An input method for entering a specific distribution order,
A physical distribution processing scenario output system including a determination means for determining which of the first or second physical distribution processing scenarios is more preferable for optimizing the scale information for the physical distribution order input by the input means.
A model generation means for generating a learned model by machine learning for defining a distribution processing scenario based on a past distribution processing scenario specified for a distribution order received in a certain period of time.
Using the learned model generated by the model generation means, a distribution processing scenario output means for outputting an optimum distribution processing scenario for a specific distribution order, and a distribution processing scenario output means.
A distribution warehouse control system including a control unit that controls and / or drives a distribution warehouse using the optimum distribution processing scenario output by the distribution processing scenario output means.