CN115239199A

CN115239199A - Distributed scheduling method based on mixed line flexible production in automobile industry

Info

Publication number: CN115239199A
Application number: CN202211029307.5A
Authority: CN
Inventors: 李成; 杨博; 郑忠斌; 陈彩莲; 关新平
Original assignee: Shanghai Jiaotong University
Current assignee: Shanghai Jiaotong University
Priority date: 2022-08-23
Filing date: 2022-08-23
Publication date: 2022-10-25

Abstract

The invention discloses a distributed scheduling method based on mixed line flexible production in the automobile industry, which relates to the field of flexible production and comprises the following steps: s101: constructing an information physical system of a factory, and interconnecting a production line, production equipment and material transportation equipment with respective digital models, wherein the information physical system is a control hub and a decision-making hub of the factory; s103: establishing a communication model of a publish-subscribe mechanism based on MQTT, wherein the MQTT message comprises timestamp information and time information corresponding to production conditions or equipment information; s105: a scheduling module is built on a central server of a factory, a distributed scheduling scheme is built in the scheduling module, and the distributed scheduling scheme determines a scheduling plan through reinforcement learning. The invention inherits the concept of flexible production, and can adjust the production schedule in time according to the sudden production condition when the condition outside the production schedule occurs in the production process, thereby increasing the robustness of the production schedule.

Description

Distributed scheduling method based on mixed line flexible production in automobile industry

Technical Field

The invention relates to the field of flexible production, in particular to a distributed scheduling method based on mixed-line flexible production in the automobile industry.

Background

With the progress and development of society, the production mode of the automobile industry is gradually changed, and since the automobile has a large demand in the market and relatively single variety in the last century, the automobile production mainly adopts the mode of a fixed production line, and nowadays, as people gradually consume automobiles from 'just needed for going out' to 'individual consumption', the types and models of automobiles are also continuously subdivided, and as a result, automobile manufacturers need to switch production plans for automobiles with more types more frequently in the production process than in the past, so that the establishment of production schedules for different orders in the automobile production process is particularly important. The following is a classified discussion about the current production modes of the whole automobile manufacturing and the characteristics of the production scheduling scheme corresponding to each mode.

At present, the whole vehicle manufacturing has three main production modes, which are respectively: workshop mode, batch production mode and mixed line production mode

Processing workshop mode: the shop mode is the production mode used by the various automotive companies prior to the application of the invention in the automotive pipeline, i.e. the assembly is performed "by hand" by skilled car workers. The mode depends heavily on the proficiency of automobile workers on the automobile production process, the production workers need to assemble a great number of parts at the same station, the workers need to master a lot of skills, the time interval between the working procedures is not fixed, and the mode mainly depends on thinking, adjusting and resting time of different workers on the process steps. When the quantity of the produced materials is large, the delivery time is long, and the transportation of the materials between different stations basically depends on a transportation vehicle in which personnel participates, so the automation level of the mode is low. Because of the high production costs and the extremely slow production speeds, the current automobile manufacturers rarely adopt this model for the production of automobiles of the type or model which meet the public needs of society, i.e. the large demand, but for the production of customized automobiles with low production volumes, such as the production of luxury branded automobiles or the production of sample automobiles, this model is particularly suitable, and the production process from the design of automobile products to the shaping of the final products needs to be continuously adjusted manually, which is a delicate process: for sample car production, a process scheme suitable for mass production and a special production line are formulated for subsequent large-scale production only if the sample car is approved by a customer; for customized luxury goods, the cost of building a special production line is extremely high due to the small demand, and instead, workshop processing with workers as the production core is more suitable, and personalized design is also conveniently introduced. Therefore, the above reasons make this model a long-standing one in the current widespread high-capacity automobile manufacturing. According to the characteristics of the processing workshop mode, the characteristics of production scheduling in the mode are mainly reflected in the scheduling of workers. The production tasks are reasonably arranged by combining the differences of different workers in the mastery degree of each process, the process steps can be executed in sequence according to time sequence, and the production scheduling is mainly characterized in that:

1. the workstations and the workers are main carriers of production tasks, and generally, production scheduling of the workers in the formulation of a scheduling plan is a main factor influencing the production schedule.

2. The time for each production task cannot be accurately determined and only a rough time range can be determined.

3. The product demand that needs scheduling is typically small and the size of the scheduling model solution is typically not large.

Batch production mode: the batch production mode is to realize the alternating batch production of automobile products of different models in time by replacing certain equipment tools, materials or workers on a certain fixed production line. Compared with a machining workshop mode, the mode has the advantages that the yield of a single automobile type can be improved, when the automobile type needs to be frequently switched to produce, a large amount of time cost is also consumed due to the fact that the production of automobiles with different models needs to be carried out on the same production line through operations of replacing equipment, materials, personnel and the like, and accordingly the yield is reduced. Therefore, a reasonable mass production plan needs to be made by calculating the conversion cost, the time length of each process in the production of different models of automobiles and the demand of different models of products in detail. Meanwhile, in consideration of reducing the production switching time of the vehicle models, the technological process of the products is fully considered at the beginning of the design of the products, the 'technological process driven product design' is emphasized, the components occupied by general parts or procedures in different vehicle models are improved, and the time cost paid in model switching is reduced. The production scheduling in this mode is characterized in that:

1. the switching time of different vehicle type production is calculated in advance, the same procedures or components in the production of different types of vehicles are fully considered, and the switching time is reduced; and determining the period of each production line in the production of each vehicle type according to the duration of each process, and then predetermining the production quantity of each batch of vehicles of each type according to the demand.

2. The scheduling and the product design are matched and supplemented.

3. The mathematical model of the scheduling problem involves a large number of parameters and is of a large scale.

Mixed line production mode: similar to a batch production mode, the mode is also used for manufacturing different vehicle type products on the same production line, but the difference is that the mode adopts a more universal production platform on the basis of the batch production mode, positioning points of a vehicle frame can be shared in various models to be produced, the production mode of production equipment is more flexible and more flexible, the production line does not need to be stopped to replace dispatching processing equipment in the switching of different vehicle type production, the vehicle type product to which the part or component to be processed currently belongs can be accurately identified, and the processing production of the current process can be carried out according to a corresponding preset program. In addition, the processing of parts by production equipment also requires positive coordination of the logistics handling system, requiring that the matching parts required by a particular vehicle model be accurately in place before a particular process is performed. At the same time, the tool is adjusted to ensure that the tool is switched 7 as flexibly as possible when producing parts of different vehicle types, without the time consumption of switching tooling tools in the batch mode being great. Therefore, highly automated and flexible logistics handling systems and production tools such as intelligent AGVs, intelligent robots are the main carriers and motives for mixed line production models. The production scheduling of the production mode is mainly characterized in that:

1. the production equipment has the characteristics of multiple processes for producing multiple vehicle types, convenience in production switching of the multiple vehicle types and the like due to the fact that execution programs of different processes of different vehicle types are stored in advance, and feasible production scheduling combination plans are various.

2. The positive cooperation of material handling and tool switching scheduling ensures that matched materials and tools are matched as much as possible when the production equipment produces a certain procedure, and the tools on the processing equipment are convenient to replace.

At present, most automobile manufacturers adopt a multi-vehicle type mixed production mode for vehicle production lines. But their degree of automation and flexibility will vary, but the production concept pursued is substantially consistent. The production mode is limited by the level of flexibility of equipment on the current production line, and is suitable for the vehicle model mixing situation with certain complexity but not too complex. At present, a centralized scheduling scheme is generally adopted for making a scheduling plan for mixed line production, and the scheme is characterized in that the production plan needs to be made again when an accident happens to actual production.

Therefore, those skilled in the art are dedicated to develop a distributed scheduling method based on mixed-line flexible production in the automobile industry.

Disclosure of Invention

In view of the above defects in the prior art, the technical problem to be solved by the present invention is how to reduce the computational complexity of the central server through a distributed scheduling plan making scheme on the basis of mixed-line flexible production, and meanwhile, when an unexpected production condition occurs, the production plan can be correspondingly adjusted according to the actual production condition, so that the making of the scheduling plan is more adaptive.

In order to achieve the aim, the invention provides a distributed scheduling method based on mixed-line flexible production in the automobile industry, which comprises the following steps:

s101: constructing an information physical system of a factory, and interconnecting a production line, production equipment and material transportation equipment with respective digital models, wherein the information physical system is a control hub and a decision-making hub of the factory;

s103: establishing a communication model of a publish-subscribe mechanism based on MQTT, wherein the MQTT message comprises timestamp information and time information corresponding to production conditions or equipment information;

s105: and constructing a scheduling module on a central server of the factory, wherein a distributed scheduling scheme is built in the scheduling module, and the distributed scheduling scheme determines a scheduling plan through reinforcement learning.

Further, the cyber-physical system in the step S101 includes a central service platform, where the central service platform includes a communication module, a digital twin module and a scheduling module, and there is data interaction between the three modules; wherein, the first and the second end of the pipe are connected with each other,

the communication module is responsible for acquiring production information of each device and storing acquired data into a database;

the digital twin module is used for carrying out digital modeling on equipment and data used for production in the factory so as to facilitate scheduling control and visual display;

and the scheduling module formulates a scheduling plan according to the production condition information of the production equipment and the material transportation equipment contained in the digital twin module.

Furthermore, the scheduling module comprises a scheduling algorithm and a scheduling sand table, wherein the scheduling algorithm is provided with an algorithm set required by the scheduling plan and an interface corresponding to the algorithm; the scheduling sand table is set as a simulation platform for production scheduling, and the scheduling plan obtained through the scheduling algorithm is simulated in the scheduling sand table.

Further, the step S101 includes the steps of:

s1011: an OPC-UA protocol client is deployed on the production line, the production equipment and the material transportation equipment, converts data on the production line and/or the production equipment into a digital model, and performs information interaction with the digital twin module according to an OPC-UA protocol;

s1012: deploying MQTT clients on the production line, the production equipment, the material transportation equipment and the detection equipment, wherein the MQTT clients publish and subscribe production information and logistics information;

s1013: deploying a first control module on the production line and a controller of the production equipment, wherein the first control module contains processing programs of different types of products, and the processing programs used by the production line in the actual production process are determined by scheduling of the scheduling plan;

s1014: and deploying a second control module on the material transportation equipment, wherein the second control module contains scheduling commands such as transportation material types, carrying capacity, path planning and the like, and the logistics transportation task of the material transportation equipment in the actual production process is determined by scheduling of the scheduling plan.

Further, the step S103 includes the following steps:

s1031: the communication module on the central service platform is set to subscribe theme messages through an MQTT client, wherein the theme messages comprise theme messages of production states of each production line, theme messages of carrying information of each material transportation device and theme messages of detection information of the detection device;

s1032: setting each communication module on the production line and the production equipment to subscribe a theme message through an MQTT client, wherein the theme message comprises: the system comprises production information and production information of products or components and scheduling strategy information of the central service platform;

s1033: and the communication module on each material transportation device is arranged to subscribe theme messages through an MQTT client, wherein the theme messages comprise the theme messages of the production state of each production line, the scheduling strategy information of the central service platform and the theme messages of the detection requirements of the detection device.

Further, the distributed scheduling scheme in step S105 includes a policy adjustment phase and a policy stabilization phase, where the policy adjustment phase and the policy stabilization phase both include the following steps:

s1051: abstracting the production line, the material transportation equipment and the detection equipment into corresponding digital scheduling objects in the scheduling module according to a digital model in the digital twin module, wherein the digital scheduling objects comprise object parameters, subscription parameters, release parameters, object models, and trigger modes and trigger conditions of the object models;

s1052: initializing fixed parameters in the production scheduling scheme: determining the total production quantity of each type of product and the components according to the production flow of different types of products and the quantity relation in the production of the components involved in the production, and determining the production value of each product or component according to the production relation;

s1053: entering the strategy adjustment stage, simulating each digital model on the central service platform, and finishing the state-behavior neural network training of the production line and the material transportation equipment;

s1054: and entering the strategy stabilization phase, distributing each digital model in the central service platform to the production line and the production equipment, and executing a scheduling scheme according to the parameters in the state-behavior neural network in the digital model.

Further, the abstraction of the digital model to the digital scheduling object in the step S1051 includes the following steps:

s10511: the abstraction of the production equipment digital model comprises the following steps: the object parameters are the types of products which can be produced by the production equipment, the subscription parameters are the production tempo of the production line in which the object parameters are located and the types of the currently produced products, the release parameters are the demand of the current production on other required components, the object model comprises a production relation model of the products, the triggering conditions of the object model are periodic triggering, and the triggering period is the numerical value of the production tempo;

s10512: the abstraction of the production line digital model comprises the following steps: the object parameters are the types and production beats of products which can be produced by the production line, the subscription parameters are the production conditions of other product or component production lines, the release parameters are the type numbers and production efficiency of the current produced products or components and on-line yield cache information, and the object model comprises a production relation model, a state-value neural network and a state-behavior neural network of the products; the subscription parameters are set as input parameters of the state-value neural network and the state-behavior neural network, the trigger mode of a production relation model in the object model is periodic trigger, the trigger period is a numerical value of a production beat, and the trigger modes of the state-behavior neural network and the state-value neural network are external trigger;

s10513: the abstraction of the material transportation equipment digital model comprises the following steps: the object parameters are the carrying amount of each product or part by the material transportation equipment and the time length required for transportation among different production equipment or production lines, the subscription parameters are production information of all the production equipment and the production lines, the release parameters are the transportation state of the material, and the object model comprises a material transportation model, the state-value neural network and the state-behavior neural network; the subscription parameters are set as input parameters of the state-value neural network and the state-behavior neural network, the triggering mode of the transportation model is internal triggering, and the triggering mode of the state-behavior neural network and the state-value neural network is external triggering;

s10514: the abstraction of the digital model of the detection device comprises the following steps: the object parameters are the types and detection duration of products which can be detected by the detection equipment, the subscription parameters are the production conditions of the production lines, the release parameters are quality information of the products or components, the object models are weight models of the production lines, and the weight models determine the production weights of the production lines which produce the same components or products.

Further, in the step S1052, each of the production lines, each of the production devices, each of the material transportation devices, and each of the detection devices are abstracted into a single digital model, and parameters in the digital model are initialized.

Further, in step S1053, the training of the state-behavior neural network includes generating training data and training the state-behavior neural network by using the training data, and specifically includes the following steps:

s10531: the production lines acquire corresponding subscription information, the state-behavior neural network determines the probability of different production behaviors of each production line in the next step, the probability is used for sampling, and the theme message of the production state of the production line is published;

s10532: the material transportation equipment acquires corresponding subscription information, the state-behavior neural network determines the probability of different transportation behaviors, samples are carried out according to the probability, and topic messages of the material transportation equipment carrying information are published;

s10533: calculating the sum of the production values of the products or parts produced in the steps of S10531 and S10531, and executing the steps of S10531 and S10531 in a circulating way until the training data meets a preset number;

s10534: calculating an intermediate variable δ = R + γ v (S', w) -v (S, w),

wherein v is an output result of the state-value neural network, w is a parameter of the state-value neural network, R is a total value of all products or parts produced by the current production behavior, γ =0.95, S is the current state, and S' is a new state;

s10535: updating parameters of a state-value neural network

Wherein β is a constant;

s10536: updating parameters of a state-behavioral neural network

Wherein, alpha is a constant, pi (·,) is an output result of the state-behavior neural network, and theta is a parameter of the state-behavior neural network;

s10537: and replacing the current state S with a new state S 'to execute the step S10534 again until the state-value neural network parameter w and the state-behavior neural network parameter theta reach a stable condition, wherein the change ranges of the stable condition parameter w and the theta value are smaller than a preset value, or the new state S' is a state of complete production.

Further, the step S1054 includes the steps of:

s10541: the central service platform distributes the digital models for each production device and each production line, and sets the initialized production weight values of each detection device to different production lines to be equal;

s10542: sampling the execution scheme of the next step by each production line, each production device and each material transportation device according to the state-behavior neural network in the digital model, executing a dynamic scheduling scheme, and putting the scheduling scheme into a training data set; for the material transport equipment, replacing the probability in the scheduling scheme by a normalized probability, wherein the normalized probability is calculated by the formula

w is the production weight;

s10543: determining the qualification rate q of each product or part on the production line through an actual detection link _n By the formula

Determining a production weight for each of the production lines; if the qualification rate of a certain production line is less than the qualified rate up to the standard, stopping the production of the production line, and overhauling the production line;

s10544: and after a period of execution time, returning to a strategy adjustment stage, and continuously training the state-behavior neural network on the basis of the existing operation parameters.

In the preferred embodiment of the invention, compared with the existing technology for making the scheduling plan of mixed line production in the whole vehicle manufacturing industry, the invention has the advantages that:

1. when the condition outside the scheduling plan occurs in the production process with the idea of flexible production, the scheduling plan can be adjusted in time according to the sudden production condition, the robustness of the scheduling plan is increased, and the production task is scheduled again without suspending the production line like a centralized scheduling scheme.

2. Compared with the existing scheme for making a centralized scheduling plan, the distributed scheduling technology is beneficial to solving the scheduling problem.

The conception, the specific structure and the technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, the features and the effects of the present invention.

Drawings

FIG. 1 is a schematic diagram of a distributed scheduling method based on mixed-line flexible production in the automotive industry according to a preferred embodiment of the present invention;

FIG. 2 is a schematic view of a prior art process for manufacturing a finished automobile;

FIG. 3 is a schematic diagram of a factory CPS framework based on mixed-line flexible production in accordance with a preferred embodiment of the present invention;

fig. 4 is a message topic involved in the MQTT publish-subscribe mechanism of the distributed scheduling method of a preferred embodiment of the present invention.

Detailed Description

The technical contents of the preferred embodiments of the present invention will be more clearly and easily understood by referring to the drawings attached to the specification. The present invention may be embodied in many different forms of embodiments and the scope of the invention is not limited to the embodiments set forth herein.

In the drawings, structurally identical elements are represented by like reference numerals, and structurally or functionally similar elements are represented by like reference numerals throughout the several views. The size and thickness of each component shown in the drawings are arbitrarily illustrated, and the present invention is not limited to the size and thickness of each component. The thickness of the components may be exaggerated where appropriate in the figures to improve clarity.

As shown in fig. 1, a schematic diagram of a distributed scheduling method based on mixed-line flexible production in the automobile industry according to an embodiment of the present invention includes the following steps:

s101: and constructing an information physical system of the factory, wherein the information physical system is a control hub and a decision-making hub of the factory, and the production line, the production equipment and the material transportation equipment are interconnected with respective digital models, as shown in fig. 3. The information physical system comprises a central service platform, wherein the central service platform comprises a communication module, a digital twin module and a scheduling module, data interaction exists among the three modules, and the specific functions of the modules are as follows:

1) The communication module is responsible for acquiring production information of each device and storing acquired data into a database;

2) The digital twin module is used for carrying out digital modeling on equipment and data for production in the factory so as to facilitate scheduling control and visual display;

3) And the scheduling module is used for making a scheduling plan according to the production condition information of the production equipment and the material transportation equipment contained in the digital twin module. The scheduling module comprises a scheduling algorithm and a scheduling sand table, wherein the scheduling algorithm is provided with an algorithm set required by a scheduling plan and an interface corresponding to the algorithm; the scheduling sand table is set as a simulation platform for production scheduling, and a scheduling plan obtained through a scheduling algorithm is simulated in the scheduling sand table.

The method comprises the following specific steps:

s1014: and deploying a second control module on the material transportation equipment, wherein the second control module comprises scheduling commands such as transportation material types, carrying capacity and path planning, and the logistics transportation task of the material transportation equipment in the actual production process is determined by scheduling the scheduling plan.

S103: and establishing a communication model of a publish-subscribe mechanism based on MQTT, wherein the MQTT message comprises timestamp information and time information corresponding to production conditions or equipment information.

The method comprises the following detailed steps:

s1032: setting the communication module on each production line and each production device to subscribe theme messages through an MQTT client, wherein the theme messages comprise: the system comprises production information and production information of products or components and scheduling strategy information of the central service platform;

The distributed scheduling scheme comprises a strategy adjusting stage and a strategy stabilizing stage, wherein each stage comprises the following steps:

the abstraction of the digital model to the digital scheduling object comprises the following steps:

s10512: the abstraction of the production line digital model comprises the following steps: the object parameters are the types and production beats of products produced by the production line, the subscription parameters are the production conditions of other product or component production lines, the release parameters are the type numbers and production efficiency of the current produced products or components and online yield cache information, and the object model comprises a production relation model, a state-value neural network and a state-behavior neural network of the products; the subscription parameters are set as input parameters of the state-value neural network and the state-behavior neural network, the trigger mode of a production relation model in the object model is periodic trigger, the trigger period is a numerical value of a production beat, and the trigger modes of the state-behavior neural network and the state-value neural network are external trigger;

s10513: the abstraction of the material transportation equipment digital model comprises the following steps: the object parameters are the carrying amount of the material transportation equipment for each product or part and the time length required by transportation between different production equipment or production lines, the subscription parameters are production information of all the production equipment and the production lines, the release parameters are the transportation state of the material, and the object model comprises a material transportation model, the state-value neural network and the state-behavior neural network; the subscription parameters are set as input parameters of the state-value neural network and the state-behavior neural network, the trigger mode of the transportation model is internal trigger, and the trigger modes of the state-behavior neural network and the state-value neural network are external trigger;

S1052: initializing fixed parameters in the production scheduling scheme: determining the total production quantity of each type of product and the components according to the production flow of different types of products and the quantity relation in the production of the components involved in the production, and determining the production value of each product or component according to the production relation. And abstracting each production line, each production device, each material transportation device and each detection device into an independent digital model, and initializing parameters in the digital models.

S1053: and entering the strategy adjusting stage, simulating each digital model on the central service platform, and finishing the state-behavior neural network training of the production line and the material transportation equipment.

The training of the state-behavior neural network comprises the steps of producing training data and training the state-behavior neural network by using the training data, wherein the specific training process comprises the following steps:

s10532: the material transportation equipment acquires corresponding subscription information, the state-behavior neural network determines the probabilities of different transportation behaviors, samples are carried out according to the probabilities, and topic messages of the information carried by the material transportation equipment are published;

s10533: calculating the sum of the production values of the products or parts produced in the steps of S10531 and S10531, and executing the steps of S10531 and S10531 in a circulating manner until the training data meets a preset quantity;

s10534: calculating an intermediate variable δ = R + γ v (S', w) -v (S, w),

wherein v is the output result of the state-value neural network, w is a parameter of the state-value neural network, R is the total value of all products or components produced by the current production behavior, γ =0.95, S is the current state, and S' is the new state;

s10535: updating parameters of a state-value neural network

Wherein β is a constant;

s10536: updating parameters of a state-behavioral neural network

Wherein, alpha is a constant, pi (·,) is the output result of the state-behavior neural network, and theta is the parameter of the state-behavior neural network;

s10537: and replacing the current state S with a new state S 'and re-executing the step S10534 until the state-value neural network parameter w and the state-behavior neural network parameter theta reach a stable condition, wherein the change range of the stable condition parameter w and the change range of the theta are smaller than a preset value, or the new state S' is a state of complete production.

The method specifically comprises the following steps:

w is the production weight;

Determining a production weight for each of the production lines; if the qualified rate of a certain production line is less than the qualified rate up to the standard, stopping the production of the production line, and overhauling the production line;

The invention provides a feasible technical support for a distributed production scheduling scheme by setting an interactive protocol of production information between cloud edges on the basis of the current popular cloud edge cooperation technology and MQTT communication technology. The invention fully utilizes the flexible and dynamic characteristics of the distributed scheduling scheme, can automatically respond through the dynamic scheduling strategy scheme in the invention when equipment failure or order insertion occurs, has high adaptivity, reduces the workload of scheduling technicians, and is beneficial to realizing the automation of production. When the condition outside the scheduling plan occurs in the production process with the idea of flexible production, the scheduling plan can be adjusted in time according to the sudden production condition, the robustness of the scheduling plan is increased, and the production task is scheduled again without suspending the production line like a centralized scheduling scheme.

As shown in fig. 2, in the process flow of manufacturing the whole automobile in the prior art, each production line produces according to its own scheduling plan, and performs subsequent related processes such as coating and detection after being assembled with the body assembly.

The following describes a distributed scheduling method based on mixed-line flexible production in the automobile industry in detail with reference to the preferred embodiment of the present invention.

A distributed scheduling method based on mixed line flexible production in the automobile industry mainly comprises three major steps, wherein each major step comprises a plurality of minor steps, and the specific steps are as follows:

the method comprises the following steps: the method comprises the following steps of constructing a plant information Physical system (CPS) which is a hub for controlling each device of a whole plant and executing decision of production tasks, wherein the CPS mainly has the following functions: 1. production scheduling, 2, uploading production information of each device and issuing an execution command, 3, performing information interaction on different devices through the platform, and 4, visually displaying the production scheduling conditions, as shown in fig. 3. The method comprises the following specific steps:

s1, building a central service platform of a factory, wherein the central service platform comprises a communication module, a digital twin module and a scheduling module, and data interaction exists among the three modules. The communication module is responsible for collecting the production information of each device and storing the data in a database so that the central server can analyze the data or other devices can inquire the production state of the devices. And meanwhile, the production information of the equipment is also used in a digital model in the digital twin module. The digital twin module is used for digitally modeling equipment and data for production in a factory so as to schedule control and visually display. The scheduling module is mainly used for making a scheduling plan according to information such as production conditions and the like contained in a digital model of production equipment and transportation equipment in the digital twin module, the scheduling module comprises two components, namely an algorithm and a sand table, and the algorithm is a set of algorithms available for the scheduling plan and an interface corresponding to the algorithm; the sand table part is a simulation platform for production scheduling, a production scheduling plan obtained through a scheduling algorithm is simulated in the sand table, particularly, under the condition of order insertion, a production plan newly formulated due to the inclusion of an order insertion product needs to be simulated and previewed in the sand table, if the constraint condition of an original order is not met, the production of the order insertion is abandoned, and if the constraint condition of the original order is met, the order insertion is accepted. The specific steps for building the central service platform of the factory are as follows:

sub1, building a digital twin module, wherein the digital twin module comprises an OPC-UA (Object Linking and Embedding) server and a visualization model tool. The production elements, namely equipment and resources in a factory are digitally modeled through OPC-UA protocol, wherein the digital modeling comprises the digital modeling of parameters and attributes of stamping, welding, painting and assembly production lines and various processing equipment on each production line and the digital modeling of parameters and attributes of logistics transportation equipment. Parameters and attributes of the production line include: the number of the production line, the number of production equipment matched with the production line, the type of parts or products produced by the production line, the output of the production line, output cache and the like; parameters and attributes of the production equipment include: production equipment number, type of used tools, type of clamps, type of processed parts, processing state, output, feeding buffer memory and the like; the parameters and attributes of the logistics transportation equipment comprise: transport equipment number, clamp type, transport state, load type, load capacity, and the like.

Sub2, building a communication module. The communication protocols to be fulfilled on the communication module include:

1. MQTT: the method is used for production information interaction among different production lines, production lines and transportation equipment. At this time, a browser and a client of the MQTT need to be deployed in a communication module of the central service platform system.

2. HTTPS: the method is used for accessing the central service platform system by an external network, and generally refers to accessing a digital twin subsystem (module), such as external intervention control, visual display and the like.

3. UDP: the method is used for transmitting a large amount of media data in industrial production, such as video and audio.

And Sub3, building a scheduling module, wherein the scheduling module is communicated with the digital twin module, can access the OPC-UA digital model, obtains parameters related to the production state of each production line, each production device and the transportation device, and can make a scheduling plan aiming at the current production condition by selecting a corresponding scheduling algorithm in a scheduling algorithm library.

S2, interconnecting different production lines, production equipment and material transportation equipment matched with the different production lines with respective digital models, namely reflecting the current production condition on a digital twin system in real time through an OPC-UA protocol, and specifically comprising the following steps:

and Sub1, deploying clients of OPC-UA protocols on different production lines and production equipment and material transportation equipment matched with the production lines, converting data on the production lines or the equipment into digital models, and performing information interaction with a digital twin system of the platform according to the OPC-UA protocols.

And Sub2, deploying MQTT clients on different production lines and matched production equipment, material transportation equipment and detection equipment thereof for publishing and subscribing production information and logistics information.

Sub3, deploying control modules on controllers of the production line and the production equipment, wherein the control modules contain processing programs of different types of products, and the specific processing program of the production line in the actual production process is determined by scheduling of a scheduling plan.

And Sub4, deploying a control module on the logistics transportation equipment, wherein the control module comprises scheduling commands of transportation material types, carrying capacity, path planning and the like, and a specific logistics transportation task in the actual production process is determined by scheduling of a scheduling plan.

Step two: a communication protocol of a publish-subscribe mechanism based on MQTT is established, and it is specified that messages of various topics discussed below all need to include timestamp information and time information corresponding to main information in the messages, i.e., production conditions or device information, and details are not described in the following step introduction, and the message topics related in the MQTT publish-subscribe mechanism are shown in fig. 4. If there is no special description, the time corresponding to the main information in the message is the time stamp time, and if there is a difference between the two, it will be mentioned in the following description. The specific implementation steps are as follows:

s1, a communication module on a central service platform is set to subscribe to messages of three themes through a client of an MQTT: subscribing production state topic information about each production line, wherein the topic format of the information is [ production line number ] - [ product or part type number ] -production information ], the content of the information is the production quantity, the production efficiency, the production cache and other information of the subscribed products or parts produced on the production line, and the topic information is issued by MQTTclient in the communication module of each production line; subscribing a theme message about each transport device, wherein the theme format of the message is ' transport device number ' -carrying information ', the content of the message is the type of the current transport product or part of the subscribed transport device and the information of the transport volume, the destination production line, the device and the like, and the theme message is issued by an MQTT client in a communication module of each transport device; the method comprises the steps of subscribing theme messages related to the detection equipment, wherein the theme format of the messages is ' product or part type number ' -detection information ', the theme messages are issued by an MQTT client in a communication module of each detection equipment, the content of the messages is the qualification rate of subscribed products or parts in a certain number N of recently produced units and a manufacturing line corresponding to unqualified products, and the number N is a preset constant.

S2, setting the communication modules on each production line and production equipment to subscribe the messages of three themes through the client of the MQTT: subscribing the yield information and the production information of the product or the component, wherein the message subject formats of the two kinds of information are respectively; "[ product or part type number ] production information" and "[ product or part type number ] stage production information". The message content of the theme of ' product or part type number ' -yield information ' comprises the total yield information and the production efficiency of subscribed products or parts, and is issued by an MQTT client in a communication module of a central service platform; the content of the theme message of ' product or part type number ' -stage production information ' comprises the production time period information of the subscribed product and the information of yield, production efficiency and the like in the time period, and is issued by the MQTT client of the communication module of each production line. Subscribing scheduling strategy information of the central service platform, wherein the subject format of the message is 'central service platform-production scheduling strategy', the content of the message is a scheduling plan instruction calculated by a scheduling module in the central service platform, and the subject message is issued by an MQTT client in a communication module of the central service platform;

s3, setting a communication module on each transport device to subscribe the messages of three themes through a client of the MQTT: subscribing to a production state subject message about each production line, wherein the message subject and content are the same as those in S1; subscribing scheduling strategy information of a central service platform, wherein the subject format of the message is 'central service platform-logistics scheduling strategy', the content of the message is a scheduling plan instruction calculated by a scheduling module in the central service platform, and the subject message is issued by an MQTT client in a communication module of the central service platform; the method comprises the steps of subscribing theme messages related to the detection equipment, wherein the theme format of the messages is ' detection equipment number ', ' type number of a product or a part to be detected ' -detection requirement ', the content of the messages is the quantity of the subscribed detection equipment required by the subscribed detection equipment for the subscribed product or part, and the theme messages are issued by MQTT clients in communication modules of the detection equipment.

Step three: constructing a scheduling module on a central server of a factory, and building a built-in distributed scheduling scheme in the scheduling module, wherein the scheme determines a scheduling plan in a reinforcement learning mode and is implemented on the premise that 1, each production device can identify the type of a part to be machined currently and the type of a product to be machined currently; 2. each device or production line can perform address resolution labeling on the produced product or component, for example, a bar code or a two-dimensional code is used as a carrier to label information such as production devices, production lines and corresponding production parameters related in the production link on the produced product or component; 3. the detection equipment can trace the production process of the product through an address resolution technology. The distributed scheduling scheme is mainly divided into two stages, wherein the first stage is a strategy adjusting stage, the second stage is a strategy stabilizing stage, and each stage comprises the following specific steps:

s1, abstracting each production line, each transportation device and each detection device into corresponding digital scheduling objects in a scheduling module according to a digital model in a digital twin, wherein the digital scheduling objects comprise four components: object parameters, subscription parameters, publication parameters, object models, and model trigger patterns and conditions. There are three types of model trigger modes, namely periodic trigger, internal event trigger and external event trigger. The period triggering is to execute the operation in the model with a certain period T, wherein T is a predetermined constant, and the triggering condition is a time interval; the internal event trigger is a mode that the object model performs operation only when the object parameter or the release parameter meets a certain condition, and is referred to as internal event trigger hereinafter; the external event trigger is a mode of executing the calculation of the object model when the subscription parameter satisfies a certain condition, and is hereinafter abbreviated as external trigger. The following steps are detailed for abstracting 4 kinds of digital models from the production equipment, the production line, the transportation equipment and the detection equipment to the digital scheduling object respectively:

sub1, abstraction of a production equipment digital model: the object parameter is the type of the product which can be produced by the equipment; the subscription parameters are the production beat of the production line and the type of the current produced product; the release parameter is the demand of the current production on other required parts; the object model comprises a production relation model of the product, the triggering condition of the model is periodic triggering, and the triggering period is a numerical value of the production rhythm.

Sub2, abstraction of production line digital model: the object parameters are the types and production beats of the products produced by the production line, and the subscription parameters are the production conditions of other product or component production lines; the release parameters are information such as the type number of the current produced product or part, the production efficiency of the current produced product or part, online yield cache and the like; the composition of the object model includes a production relationship model of the product, a state-value neural network, and a state-behavior (or state-action) neural network. The subscription parameters serve as input layers for both the state-value neural network and the state-behavior neural network. For the state-behavior neural network, the kind of the produced product in the parameters and different production efficiencies are released as each neuron of the output layer of the neural network, and the output value of the neuron of the output layer represents the probability of each production case. For a state-value neural network, the output is a neuron, and the initialization of the neural network is 0. The triggering mode of the production relation model is periodic triggering, the triggering period is a numerical value of a production beat, the triggering mode of the state-behavior neural network is external triggering, and the triggering modes of the state-behavior neural network and the state-value neural network are external triggering.

Sub3, abstraction of the transportation equipment digital model: the object parameters are the carrying amount of the equipment for each product or part and the time length required for transportation between different production equipment or production lines; the subscription parameters are production information of all production lines and production equipment; the release parameter is the transportation condition of the material; the object model comprises a transportation model of the material, a state-value neural network and a state-behavior neural network. The subscription parameters serve as input layers for both the state-value neural network and the state-behavior neural network. For the state-behavioral neural network, each transport state case in the release parameters, i.e., information (production line L, product P, target device D) for transporting a certain product from a certain production line to a certain production device, is respectively taken as each neuron of the output layer of the state-behavioral neural network, and the output value of the neuron of the output layer represents the probability of each transport case. For a state-value neural network, the output is a neuron, and the initialization of the neural network is 0. The triggering mode of the transportation model is internal triggering, and the triggering mode of the state-behavior neural network and the state-value neural network is external triggering.

Sub4, abstraction of a digital model of the detection equipment: the object parameters are the types and detection duration of products which can be detected by the equipment; with subscription parameters for individual production linesProduction conditions; the release parameter is quality information of the product or the component; the object model is a weight model of each production line, the weight model determines the production weight of the production lines producing the same parts or products, and the calculation formula is

Wherein q is _n Is the qualification rate q of n number production lines ₀ The qualified rate of the product reaches the standard. Q of mass _n The information is not determined by the model in the present invention, but by the actual detection process. However, during initialization, the production weight of each detection device to different production lines is equal and is 1.

S2, initializing fixed parameters in the production scheduling scheme: the production flow of the different models of products and the quantity relation involved in the production of the components are used to determine the products of each model and the total production quantity of the components, and the production value V of each component or product is determined according to the production relation. If this is the case, the order is accepted and the order product is blended with the incomplete products or parts in the existing order to determine the total amount of each part currently still to be produced. Abstracting each production line, each production device, each transportation device and each detection device into a single digital model, and initializing parameters in the model, wherein the model relates to a state-behavior neural network, and the parameters are randomly initialized.

And then entering a first stage, namely S3, in the distributed scheduling scheme, because the qualification rate of product detection cannot be known in the first stage, in a strategy adjustment stage, a digital scheduling object of the detection equipment does not participate in simulation.

S3, entering a first stage, namely a strategy adjustment stage, and simulating each digital model on a central service platform, wherein the specific steps of the stage are as follows:

sub1, production training data:

SubSub1, the production line obtains corresponding subscription information, the state-behavior neural network respectively determines the probability of different production behaviors of each production line in the next step, and the probability is used for sampling and publishing production information subject information.

And SubSub2, the transportation equipment acquires corresponding subscription information, determines the probabilities of different transportation behaviors by the state-behavior neural network respectively, samples according to the probabilities and publishes the information carrying subject message.

SubSub3, calculating the sum I of the production values of all the products or parts produced in the current two steps for the simulation process of the last two steps, and circulating SubSub1 and SubSub2 until the training data meets a certain amount.

Sub2, training the state-behavior neural network of the production line and the transportation equipment through the produced training data. The training process is as follows:

SubSub1, calculating an intermediate variable δ = R + γ v (S', w) -v (S, w), wherein v is an output result of the state-value neural network, w is a parameter of the neural network, R is a total value of all products or components produced by the current production behavior, and γ is a preset constant and is generally 0.95.

Subsub2, updating parameters of a state-value neural network

Beta is a predetermined constant.

Subsub3, updating of parameters of the State-behavioral neural network

Alpha is a preset constant, pi (·,) is the output result of the state-action neural network, and theta is the parameter of the neural network.

SubSub4, replacing the current state S with a new state S' and then SubSub1 again until the parameters of the two neural networks reach stable conditions, wherein the stable conditions are two types: 1. the variation range of the parameters of the models of the two neural networks along with training is smaller than a preset value; 2. the new state S' is a fully produced state.

And S4, entering a second stage, namely a strategy stabilization stage, distributing each digital model in the central platform to each production line and equipment, and executing a scheduling scheme according to the parameters in the state-behavior neural network in the digital model. Because the actual production operation comprises a detection link, the production data in the scheduling execution process is still put back into the training data set, and the training is continued on the basis of the reappearance of some model parameters after returning to the first stage every time. The specific steps in this stage are as follows:

sub1, the central service platform distributes digital models for each device and each production line, and the production weight of each detection device to different production lines is initialized to be equal and is 1.

Sub2, each production line, each production device and each transport device sample the execution scheme of the next step according to the state-behavior neural network in the digital model, then execute the dynamic scheduling scheme and put the execution scheme of the step into the training data set. In the step, for the transportation equipment, the probability of different transportation conditions obtained by a state-behavior neural network in an original digital model is multiplied by the production weight of the related production line on the detection equipment, and then the probability normalization is carried out again

w is the production weight.

Sub3, determining the qualification rate q of each product or part on the production line through the actual detection link _n By the formula

The production weight of each production line is determined. And if the qualification rate of a certain production line is less than the qualified rate up to the standard, stopping the production of the production line and overhauling the production line.

And Sub4, after a period of execution time, returning to the first stage, and continuing training on the basis of the existing parameters.

The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.

Claims

1. A distributed scheduling method based on mixed line flexible production in the automobile industry is characterized by comprising the following steps:

2. The method of claim 1, wherein the cyber-physical system in the step S101 includes a central service platform, the central service platform includes a communication module, a digital twin module and a scheduling module, and there is data interaction among the three modules; wherein the content of the first and second substances,

and the scheduling module is used for making a scheduling plan according to the production condition information of the production equipment and the material transportation equipment contained in the digital twin module.

3. The method of claim 2, wherein the scheduling module comprises a scheduling algorithm and a scheduling sandbox, the scheduling algorithm being provided with a set of algorithms required by the scheduling plan and an interface corresponding to the algorithms; the scheduling sand table is set as a simulation platform for production scheduling, and the scheduling plan obtained through the scheduling algorithm is simulated in the scheduling sand table.

4. The method of claim 1, wherein the step S101 comprises the steps of:

5. The method of claim 1, wherein the step S103 comprises the steps of:

s1033: and setting the communication module on each material transportation device to subscribe theme messages through an MQTT client, wherein the theme messages comprise the theme messages of the production state of each production line, the scheduling strategy information of the central service platform and the theme messages of the detection requirement of the detection device.

6. The method of claim 1, wherein the distributed scheduling scheme in step S105 comprises a policy adjustment phase and a policy stabilization phase, each of which comprises the steps of:

7. The method of claim 6, wherein the abstraction of the digital model to the digital scheduling object in the step S1051 comprises the steps of:

s10511: the abstraction of the digital model of the production equipment comprises the following steps: the object parameters are the types of products which can be produced by the production equipment, the subscription parameters are the production tempo of the production line in which the object parameters are located and the types of the currently produced products, the release parameters are the demand of the current production on other required components, the object model comprises a production relation model of the products, the triggering conditions of the object model are periodic triggering, and the triggering period is the numerical value of the production tempo;

8. The method according to claim 6, wherein in step S1052, each of the production lines, each of the production devices, each of the material transporting devices, and each of the detecting devices are abstracted into a single digital model, and parameters in the digital model are initialized.

9. The method of claim 6, wherein the training of the state-behavior neural network in step S1053 includes producing training data and training the state-behavior neural network using the training data, and specifically includes the steps of:

s10531: the production lines acquire corresponding subscription information, the state-behavior neural network determines the probability of different production behaviors of each production line in the next step, the probability is used for sampling, and the theme message of the production state of the production lines is published;

s10534: calculating an intermediate variable δ = R + γ upsilon (S', w) -upsilon (S, w),

wherein ν is an output result of the state-value neural network, w is a parameter of the state-value neural network, R is a total value of all products or parts produced by the current production behavior, γ =0.95, S is the current state, and S' is a new state;

s10535: updating parameters of a state-value neural network

Wherein β is a constant;

s10536: updating parameters of a state-behavioral neural network

10. The method of claim 6, wherein the step S1054 comprises the steps of:

s10542: sampling the execution scheme of the next step by each production line, each production device and each material transportation device according to the state-behavior neural network in the digital model, executing a dynamic scheduling scheme, and putting the scheduling scheme into a training data set; replacing in the scheduling scheme for the material handling equipment using normalized probabilitiesIs calculated by the formula of normalized probability

w is the production weight;