Theory and application of digital twin model construction

Source: Computer Integrated Manufacturing System

Authors: Tao Fei Zhang He Qi Qinglin Xu Jun Sun Zheng Hu Tianliang Liu Xiaojun Liu Tingyu Guan Juntao Chen Changyu Meng Fanwei Zhang Chenyuan Li Zhiyuan Wei Yongli Zhu Minghao Xiao Bin

Summary

As an important enabling way to realize digital transformation and promote intelligent upgrading, digital twin has always attracted the attention of all walks of life, and has moved from theoretical research to the stage of practical application. Digital twins are driven by multi-dimensional virtual models and fusion data, through virtual-real closed-loop interaction, to achieve monitoring, simulation, prediction, optimization and other practical functional services and application requirements, among which the construction of digital twin models is the premise of realizing the application of digital twins. Aiming at the problem of how to build a digital twin model, the construction criterion of "four modernizations, four functions and eight uses" of the digital twin model is first proposed. Based on the proposed construction criteria, a set of digital twin model construction theoretical system was explored and established from six aspects of "construction-group-integration-verification-calibration-management". And based on the proposed digital twin model construction criteria and theoretical system, combined with the national key research and development project "theory and method of precise modeling of intelligent production process based on digital twin" undertaken by the author's team, taking the digital twin workshop as an example, from the workshop elements The theory and technology of digital twin workshop model construction have been researched and practiced in three aspects: solid modeling, dynamic modeling of production process, and simulation modeling of production system. The related work is expected to provide a theoretical reference for the construction and further application of digital twins.

Key words

digital twin; intelligent manufacturing; digital twin workshop; modeling; construction criteria; theoretical system

1 From model to digital twin

Models are an important element in manufacturing activities, and they take on different forms and play different roles in different historical stages and under different technical backgrounds. Human beings have been using "models" to make bronze wares since the Bronze Age. For example, the "block model method" and "lost wax method" adopted in bronze casting in the Shang and Zhou dynasties in my country were based on models. The "block model method" and "lost wax method" first select pottery, wood, bamboo, bone, stone, wax and other materials to make the "physical model" of the bronze ware, and then make a mold on the basis of the model, and through Molten copper is poured into the cavity, solidified and cooled to obtain a bronze caster. Similarly, the style Lei family responsible for imperial buildings (such as palaces, imperial mausoleums, gardens, etc.) in the Qing Dynasty used the "hot sample" (i.e. "real model") of the building to turn the design into a three-dimensional miniature landscape, so that the building can be understood in advance effect, and then guide the actual construction. These "hot samples" made at a scale of 1/100 or 1/200 before the actual building starts not only show the appearance of the building in appearance, but also reflect the details of the building's platform, tile roof, pillars, doors and windows, etc. internal structure. In addition, "mock-ups" of actual physical objects can replace some of the functions of their prototypes, in addition to assisting in manufacturing. For example, the famous terracotta warriors and horses in the Mausoleum of Qin Shihuang replaced living people and were buried with Qin Shihuang. In order to truly reproduce the mental outlook of the soldiers of the Qin Army, these terracotta warriors were rendered very realistically by the craftsmen with superb skills. The faces, eyes, expressions, ages, etc. are different and vivid. In addition, during the Three Kingdoms period, Zhuge Liang invented the transportation tool "Wooden Niu Liuma" to transport grain to the 100,000 troops of the Shu Han. It has the functions and functions of real cattle and horses, thus replacing real cattle and horses for grain transportation.

The above-mentioned "physical model" can realize the duplication of corresponding physical objects or functions, but this type of model has a certain degree of time and space limitations. For example, on the time scale, the physical model mainly reproduces the appearance or structure statically, which cannot fully reflect the changing characteristics of physical objects over time; on the spatial scale, for large scenes (such as the entire city, the entire park), internal structures Complex (such as engine) physical objects, such physical models are difficult to fully describe. With the maturity and popularization of computer, information, network communication and other technologies, people can use digital technology to break through the limitations of time and space and establish a "digital model" of physical objects to solve the above problems. For example, the use of computer graphics technology, virtual reality and augmented reality technology has realized the creation of digital Yuanmingyuan (that is, the digital model of Yuanmingyuan) in the virtual world, thereby recreating the original historical appearance of Yuanmingyuan. In addition, using holographic imaging technology to resurrect the singing performance of the deceased singer on the stage, the audience can not only see the digital virtual singer with the same appearance as the singer, but also hear the singing voice exactly the same as the singer, realizing the virtual reality of the deceased. resurrection.

Whether it is the above-mentioned "physical model" for auxiliary manufacturing and partial function replacement, or the "digital model" for digital display, the description of physical objects on multi-dimensional and multi-spatial scales is not enough; in addition, their work or operation process is relatively independent , lacking dynamic interaction with corresponding physical objects. With the further development and in-depth application of the new generation of information technology, people's ever-increasing actual needs of industry and life put forward requirements for the model to be able to interact with physical objects. People want to know what is the space-time of different scales in the physical world, what is happening now, and what will happen in the future? So as to predict possible problems and formulate corresponding measures. In this context, the digital twin came into being and caused profound industrial changes [1]. Physical entities and their corresponding virtual models, data, connections, and services are the core components of digital twins [2]. Driven by multi-dimensional virtual models and fusion data, as well as the interaction between physical objects and virtual models, digital twins can describe the multi-dimensional attributes of physical objects, describe the actual behavior and state of physical objects, and analyze the future development trend of physical objects. Object monitoring, simulation, prediction, optimization and other practical functional services and application requirements [3], and to a certain extent achieve the symbiosis of physical objects and virtual models [4].

The model is an important part of the digital twin and an important prerequisite for realizing the function of the digital twin. However, how to build a digital twin model currently lacks general guidelines and theoretical systems for reference and guidance. In response to this problem, based on the previous related work, this paper proposes a set of digital twin model construction criteria and theoretical system. Based on the proposed digital twin model construction criteria and theoretical system, combined with the national key research and development project "theory and method of precise modeling of intelligent production process based on digital twin" undertaken by the team, taking the digital twin workshop as the object [5], in the workshop Based on the analysis of the current status of modeling research, the method and theory of digital twin workshop model construction are discussed from the aspects of digital twin workshop entity element modeling, data-driven production process dynamic modeling, and production system simulation modeling. Finally, the relevant theories and methods proposed in this paper are initially verified by taking the new energy vehicle power battery production workshop as an example.

2 Guidelines for building digital twin models

The author's team proposed a digital twin five-dimensional model in the early stage, including physical entities, virtual models, services, twin data, and the connection and interaction between them [5]. The digital twin model studied in this paper refers to the virtual model part of the digital twin five-dimensional model. Its main function is to describe and describe physical entities or all elements of complex systems in multiple dimensions, in multiple temporal and spatial scales, and in multiple domains. In order to make the digital twin model construction process evidence-based, this paper proposes a set of digital twin model "four modernizations and four usability" construction guidelines, as shown in Figure 1. Oriented to meet actual business needs and solve specific problems, the guidelines aim at "eight uses" (usable, general, quick, easy to use, combined, combined, flexible, and easy to use), and propose the digital twin model "four modernizations". " (precision, standardization, lightweight, visualization) requirements, and the "four can" (interactive, fused, reconfigurable, and evolvable) requirements during its operation and operation.

(1) Precision

The precision of the digital twin model means that the model can not only accurately describe and describe the physical entity or system statically, but also make the dynamic output of the model consistent with the actual or expected changes over time. The precision criterion of digital twin modeling is to ensure that the digital twin model built is accurate, accurate, credible, and usable, so as to meet the validity requirements of the digital twin model. An accurate digital twin model is an important prerequisite for the correct functioning of the digital twin. Taking the digital twin workshop as an example, an accurate digital twin model can fundamentally prevent the transmission and accumulation of model errors in the process of building a digital twin workshop, thereby effectively avoiding the error caused by iterative amplification of model errors during the operation of the digital twin workshop. Serious Problem.

(2) Standardization

The standardization of digital twin models refers to the standardization and unification of model definition, coding strategy, development process, data interface, communication protocol, solution method, model service packaging and use, etc. The standardization criteria for digital twin modeling are to ensure the consistency of model integration, model data exchange, model information identification, and model maintenance, so that digital twin models built for different elements and objects in different industries and fields are easy to analyze and can be analyzed. Reusable and compatible with each other, so as to further meet its versatility requirements on the basis of ensuring the validity of the digital twin model. Taking the digital twin workshop as an example, the standard digital twin model can not only reduce the generation of redundant models and heterogeneous models when modeling for different physical workshops, but also significantly reduce the difficulty of unified and integrated management of the digital twin workshop model.

(3) Lightweight

The lightweight of the digital twin model refers to the simplification of the model in terms of geometric description, carrying information, and construction logic under the premise of satisfying the loss of main information, model accuracy, and use functions. The lightweight criterion of digital twin modeling is to further meet the efficiency requirements of digital twin modeling and model operation for complex systems through the rapid use of digital twin models on the basis of availability and generality of digital twin models. Taking the digital twin workshop as an example, the lightweight digital twin model realizes a realistic description of the physical workshop based on relatively few parameters and variables, which not only facilitates the rapid modeling of the digital twin workshop, but also effectively reduces the transmission time of digital twin model parameters, Accelerate the running speed of the digital twin model, thereby improving the timeliness of decision-making based on online simulation in the digital twin workshop.

(4) Visualization

The visualization of the digital twin model means that the digital twin model can be presented to the user in an intuitive and visible form during the process of construction, use, and management, so as to facilitate the in-depth interaction between the user and the model. The visualization criterion of digital twin modeling is to make the constructed accurate, standard and lightweight digital twin model more readable and easy to use, and to meet the intuitive requirements of digital twin model. Taking the digital twin workshop as an example, the digital twin workshop model is composed of multi-element, multi-dimensional, multi-field, and multi-scale models. The visualized digital twin model can display the structure and evolution of the digital twin workshop model in a vivid and vivid way. The coupling relationship between the process, parameter details and its sub-models can effectively support the efficient analysis of the model and the visual operation and maintenance control of the digital twin workshop.

(5) Interactive

The interactive digital twin model means that different models and between models and other elements can exchange data and instructions through compatible interfaces, and realize model collaboration based on entity-model-data combination. The interactive criterion of digital twin modeling is to eliminate discretely distributed information islands in the system and meet the connectivity requirements for complex system modeling. Taking the digital twin workshop as an example, the digital twin workshop model can interact with the element entities in the physical workshop, which can effectively connect the physical workshop and the virtual workshop, and realize mutual control and synchronous mapping between virtual and real; on this basis, the interactive , can effectively connect the entire digital twin workshop, and realize model collaboration through model parameter sharing and knowledge complementarity; at the same time, the digital twin model and twin data can interact, and can also realize efficient data collection and transmission oriented by model operation requirements and data-driven model parameter automation. renew.

(6) can be integrated

The fusion of digital twin models means that multiple or multiple digital twin models can be integrated into a whole based on the relationship, that is, the mechanism model, model data, data characteristics and model-based decision-making can be effectively integrated. The fusion criterion of digital twin modeling is to analyze and describe complex systems more comprehensively, thoroughly, and objectively, and to meet the overall requirements for complex system modeling under the premise of system connectivity. Taking the digital twin workshop as an example, through the fusion of multi-dimensional models, the combination of multiple models, the association of multi-type models and the collaboration of multi-level models, the digital twin workshop can be represented as a unified whole, thereby generating and accumulating virtual and real data during its operation. Scale fusion data, to achieve global decision-making and optimization based on fusion models and fusion data, to help digital twin workshops operate safer and more efficiently.

(7) Reconfigurable

The reconfigurability of the digital twin model means that the model can face different application environments, and quickly meet new application requirements by flexibly changing its own structure, parameter configuration, and relationship with other models. The reconfigurable criterion of digital twin modeling is to avoid the difficulty of adapting the assembled and fused digital twin model to the dynamically changing environment, and to meet the flexibility requirements of complex system models by using the model flexibly. Taking the digital twin workshop as an example, when enterprises use the digital twin workshop for production operations, they need to consider objective requirements such as production equipment replacement, process route changes, production technology improvement, workshop capacity increase, and new product production, as well as equipment failures, personnel fatigue, For uncertain events such as environmental fluctuations, the digital twin model can be reconfigured to give the digital twin workshop the ability to be scalable, configurable, and schedulable, which improves the flexibility of the digital twin workshop and meets the urgent needs of enterprises to improve their competitiveness in the dynamic market .

(8) Evolvable

The evolution of the digital twin model means that the model can update and evolve the model function as the physical entity or system changes, and perform continuous performance optimization over time. The evolutionary criterion of digital twin modeling is to realize the self-correction and self-optimization of the model based on the static data of the whole life cycle of the model and the dynamic data of the model operation process on the basis of the above-mentioned criterion, so as to make the original model more and more easy to use , so as to meet the intelligence requirements of equipment and complex systems. Taking the digital twin workshop as an example, a large amount of real-time twin data will be generated and accumulated during the operation of the digital twin workshop. Iterative calculation based on real data in the virtual workshop will enable the model to iteratively update following changes in the physical workshop, and make the digital twin workshop Obtain continuously optimized decision-making and evaluation capabilities. At the same time, knowledge mining and knowledge accumulation based on effective data can continuously improve the intelligence of the digital twin workshop.

3 Theoretical system of digital twin model construction

A digital twin model is a digital representation of a real-world entity or system, which can be used to understand, predict, optimize, and control a real entity or system. Therefore, the construction of a digital twin model is the basis for realizing model-driven. The digital twin model construction is to realize the digital modeling of the attributes, methods, behaviors and other characteristics of physical entities and processes in the digital space. Model construction can be multi-dimensional "geometry-physics-behavior-rules" or multi-field "mechanical-electrical-hydraulic". From the perspective of work granularity or level, the digital twin model is not only the modeling of the basic unit model, but also needs to realize the construction of more complex object models through model assembly from the spatial dimension, and realize the various aspects of complex physical objects through model fusion from the perspective of multi-field and multi-discipline. A comprehensive characterization of domain characteristics. In order to ensure the correctness and effectiveness of the digital twin model, it is necessary to verify the constructed and assembled or fused model to verify whether the model description and the state or characteristics of the physical object are correct. If the model verification results do not meet the requirements, it is necessary to make the model closer to the actual operation or use state of the physical object through model calibration to ensure the accuracy of the model. In addition, model management is also necessary to facilitate operations such as addition, deletion, modification, query, and user use of the digital twin model, as well as the use of model verification or correction information. Therefore, this paper proposes a theoretical system of digital twin model construction including model construction, model assembly, model fusion, model verification, model correction, and model management, as shown in Figure 2.

3.1    Construction: Model Construction

Model construction refers to the construction of a model of its basic unit for a physical object. The digital twin model can be constructed from two aspects: multi-domain model construction and "geometry-physics-behavior-rule" multi-dimensional model construction. The "geometry-physics-behavior-rule" model can describe the geometric characteristics, physical characteristics, behavior coupling relationship and evolution law of physical objects; the multi-domain model can comprehensively describe the physical objects by building models of various fields involved in physical objects. Characteristics in various fields such as thermal science and mechanics. Through multi-dimensional model construction and multi-domain model construction, the precise construction of the digital twin model is realized. Ideally, the digital twin model should cover multi-dimensional and multi-domain models, so as to achieve a comprehensive and realistic characterization and description of physical objects. However, from the perspective of application, the digital twin model does not necessarily need to cover all dimensions and fields. At this time, it can be adjusted according to actual needs and actual objects, that is, to build models of some fields and some dimensions [6].

Group 3.2    : Model Assembly

When the model building object is relatively complex, it is necessary to solve the problem of how to go from a simple model to a complex model. Digital twin model assembly is the process of realizing the digital twin model from the unit level model to the system level model and then to the complex system level model from the spatial dimension. The realization of digital twin model assembly mainly includes the following steps: first, it is necessary to build the hierarchical relationship of the model and clarify the assembly sequence of the model to avoid difficult assembly; second, it is necessary to add appropriate space constraints during the assembly process, different levels There are certain differences in the spatial constraints that need to be paid attention to and added in the model. For example, in the model assembly process from parts to components to equipment, it is necessary to build and add constraints such as angle constraints, contact constraints, and offset constraints between parts. In the model assembly process from equipment to production line to workshop, it is necessary to build and add the spatial layout relationship between equipment and the spatial constraint relationship between production lines; finally, the model assembly is realized based on the constructed constraint relationship and model assembly sequence.

3.3   Fusion: model fusion

Model fusion is aimed at the construction of some system-level or complex system-level twin models. Model assembly in spatial dimensions cannot meet the needs of describing physical objects, and further model fusion is required, that is, to achieve fusion between models in different disciplines and different domains. In order to realize the fusion between models, it is necessary to construct the coupling relationship between the models and clarify the one-way or two-way coupling methods between different domain models. For different objects, there are certain differences in the areas of focus of the model fusion. Taking the CNC machine tool in the workshop as an example, the CNC machine tool involves multiple subsystems such as hydraulic system, electrical system, and mechanical system, and there is a coupling relationship between different systems. Fusion of liquid multi-domain models.

3.4    Validation: Model Validation

After model construction, assembly or fusion, the model needs to be verified to ensure the correctness and effectiveness of the model. Model verification is to check whether the output of the model is consistent with the output of the physical object for different requirements. In order to ensure the accuracy of the constructed model, the unit-level model is first verified after construction to ensure the validity of the basic unit model. In addition, since the model may introduce new errors during the assembly or fusion process, the assembled or fused model is not accurate enough. Therefore, in order to ensure the ability of the digital twin assembly and fusion model to accurately describe physical objects, it is necessary to further verify the assembled or fusion model on the basis of ensuring that the basic unit model is high-fidelity. If the model verification results meet the requirements, the model can be further applied. If the model verification results cannot meet the requirements, model calibration is required. Model verification and calibration is an iterative process, that is, the corrected model needs to be verified again until it meets the requirements of use or application.

3.5    Calibration: Model Calibration

Model correction means that there is a certain deviation between the verification results of the model verification and the physical object, and when the requirements cannot be met, the model parameters need to be corrected to make the model closer to the actual state or characteristics of the physical object. Model calibration mainly includes two steps: (1) Selection of model calibration parameters. Reasonable selection of calibration parameters is one of the important factors to effectively improve calibration efficiency. The selection of parameters mainly follows the following principles: ①The selected calibration parameters should have a strong correlation with the target performance parameters; ②The number of calibration parameters should be selected appropriately; ③The upper and lower limits of the calibration parameters should be set reasonably. The combination of different calibration parameters will have a certain impact on the model calibration process. (2) Correct the selected parameters. After the calibration parameters are determined, the objective function needs to be constructed reasonably. The objective function is that the output result of the corrected model is as close as possible to the physical result. Based on the objective function, an appropriate method is selected to realize the iterative calibration of the model parameters. Through model calibration, the accuracy of the model can be guaranteed, and it can be better adapted to different application requirements, conditions and scenarios.

3.6    Pipes: Model Management

Model management refers to providing convenient services for users through reasonable classification, storage and management of digital twin models and related information on the basis of model assembly fusion, verification and correction. In order to provide users with services for quickly searching, building, and using digital twin models, model management needs to have functions such as multi-dimensional model/multi-domain model management, model knowledge base management, multi-dimensional visual display, and operation operations, and supports model preview, filtering, and search operations. ;In order to support users to quickly apply the model to different scenarios, it is necessary to manage the data generated during the verification and calibration process of the model, including verification information such as verification objects, verification features, and verification results, as well as calibration objects, calibration parameters, and calibration results. Such correction information will help the model to be applied in different scenarios and guide the construction of subsequent related models.

Model construction, model assembly, model fusion, model verification, model calibration, and model management are the six major components of the digital twin model construction system, but in the actual construction process of the digital twin model, it may not be necessary to include all of these six processes. It needs to be adjusted accordingly according to the actual application requirements. For example, it is not necessary to assemble and fuse models in order to visualize a part.

4 Construction of digital twin workshop model

4.1    Research Status of Workshop Modeling

As the basic unit of enterprise production, the workshop's efficient operation and management is an important guarantee for improving quality and efficiency. In order to realize the optimization of workshop layout, process flow optimization design, intelligent decision-making and scheduling, many scholars have carried out related research on the construction of workshop virtual model. The current workshop modeling research is mainly carried out from the following three aspects: (1) describe what the workshop is—that is, model the production factors such as "human-machine-thing-environment" in the workshop; (2) describe what the workshop is doing What—that is, real-time dynamic modeling of the workshop production process; (3) predict what will happen in the workshop—that is, simulate and model the workshop production system. The specific summary and analysis are as follows:

4.2    Precise Construction Theory and Technical System of Digital Twin Workshop Model

In view of the incomplete element model, unintegrated process data, and inaccurate production system simulation in the process of building the workshop model, according to the construction principle of the digital twin model "four modernizations, four functions" proposed in this paper, the digital twin of the author's team in the early stage Based on the research work on workshop key technologies [4] and digital twin enabling technology [42], this paper further studies and proposes a theory and technical system for the precise construction of digital twin workshop models.

As shown in Figure 3, the theoretical and technical system consists of three parts: production factor entity modeling, production process dynamic modeling and production system simulation modeling. The entity modeling of production factors includes the theory of multi-dimensional and multi-domain model construction of workshop elements and related technologies (such as knowledge map technology, 3D scanning modeling technology, finite element technology, etc.), multi-level factor model assembly theory and related technologies, multi-dimensional model fusion theory and Related technologies, element model verification theory and technology, element model correction theory and technology, and model management technology; dynamic modeling of production process includes edge-cloud collaborative data collection theory and related technologies (such as edge computing technology, 5G transmission technology, etc.), workshop process Model construction related technologies and production process model self-iteration related theories and technologies; production system simulation modeling technologies include constraints construction related technologies, production system simulation model construction and update related technologies, and production system simulation model reduction and solution related theories and technologies (such as multi-agent technology, Krylov subspace projection technology, etc.). These theories and technologies together support and guarantee the accuracy of the construction of the digital twin model of the workshop.

4.3    Entity modeling of all elements in the digital twin workshop

Aiming at how to construct an accurate digital twin entity model covering all elements of the workshop, multi-dimensional, and multi-scale, to achieve accurate description of production elements, from the construction of all-element multi-field/multi-dimensional digital twin models, multi-scale twin model assembly and fusion, digital The virtual-actual consistency verification of the twin model describes the solution or process in three aspects, as shown in Figure 4.

 4.3.1 Construction of multi-domain/multi-dimensional digital twin model of all elements of the workshop

This paper proposes a new silicon-based SPAD device structure, which not only forms a main avalanche region between the deep N well and the P- epitaxial layer, but also forms two symmetrical secondary avalanche ring regions in the deep N well. The main avalanche region has a deep depth and a large area, which can improve the detection efficiency of near-infrared short-wavelength photons; the two symmetrical secondary avalanche ring regions cover a wide range in the deep N-well, which can expand the photon response band of the device. TCAD simulation results show that compared with the traditional P+/Nwell structure, the photon detection efficiency of the new SPAD device structure is increased by 5 times at 850nm, and higher photon detection efficiency is obtained in the wide spectral range of 300nm-1000nm. In addition, the dark count rate of the device is less affected by the BTBT effect, and a lower DCR is obtained below 20 °C, which greatly improves the overall performance of the device.

The construction of the multi-domain/multi-dimensional digital twin model of all elements of the workshop includes the construction of the mechanical/electrical/hydraulic multi-domain basic model and the construction of the "geometry-physics-behavior-rule" multi-dimensional twin model. (1) In terms of multi-dimensional model construction, for the geometric dimension, based on the product design geometric feature parameters and other information, construct an expandable geometric model covering heterogeneous elements of equipment; for the physical dimension, construct according to the physical characteristics of the equipment such as material properties and physical parameters Physical model; for the behavior dimension, based on the behavioral coupling relationship between components, construct a behavior and response model that describes the behavior characteristics of the equipment; for the rule dimension, describe the rules and logic models of equipment operation and evolution rules based on the XML language to form a coverage " Geometry-Physics-Behavior-Rule" digital twin basic model of multi-dimensional features [5]. (2) In terms of multi-domain model construction, first analyze the domains and related mechanisms included in the workshop equipment, and use Modelica (multi-domain unified modeling language) to express mathematical equations for meta-models in multiple domains such as mechanical/electrical/hydraulic, In this way, components, scenarios, processes, and process models oriented to the fields of electromechanical/electrical/hydraulic are formed. The multi-dimensional model and multi-domain model of production factors provide a model basis for further model assembly and fusion to form a digital twin model of the production system.

 4.3.2 Assembly and Fusion of Multi-Scale Twin Model of All Factors in Workshop

From the prior distribution of θ, it is obvious that it is a martingale about {Wt, t≥0}, and about {Bt, t≥0}. And when T is bounded, according to isometric [14], it can be known that it is a square integrable martingale. Similarly, it is a square integrable martingale, and it can be obtained from literature [15] and the law of large numbers

The workshop process is dynamic, so the production process model needs to be iteratively updated according to the actual production operation. The dynamic iterative update of the production process model includes two processes: workshop model dynamic self-awareness and workshop model dynamic self-iteration.

It is said that if a child goes home and tells his parents that he likes the teacher of a certain subject, then the parents hardly have to worry about the child failing to learn the subject. I think I am lucky, no matter middle school or university, I respect and love my math teacher. Although I have almost completely forgotten the mathematics class now, I firmly believe that mathematics must have given me unspoken enlightenment in life and work. I also miss the campus, the classroom, and the teachers who inspired me to study and improve my studies, taught me my work and solved my doubts.

The constructed production factor model and production system model need to verify the consistency of virtual reality to ensure the validity, correctness and accuracy of the model. For the digital twin model of production factors, the virtual-real consistency verification of the digital twin model is carried out from the four dimensions of geometry-physics-behavior-rule. For the geometric model, the geometric features for verification are determined on the basis of analyzing the geometric features and topological structure relationship of the geometric model, and the measurement results of the key geometric features are obtained through the selection of the number of measuring points, the measurement layout, and the planning of the measurement path, and then the tolerance value, feature Evaluate the geometric model verification results of information such as name and consistency; for the physical model, identify the anchor point of the electromechanical component from the data source or sensor of the manufacturing unit, and obtain the verification result of the physical model based on the comparative analysis of the anchor point change; for the behavior model , construct the timing diagram reflecting the timing characteristics of the behavior and the state diagram of the behavior response characteristics, and use the formal verification method to obtain the verification results of the behavior model; for the rule model, use the model verification method based on data incentives, and compare the rule model Drive the response and the actual response to get the verification results of the rule model. According to the verification results of the four dimensions, the comprehensive verification results of the digital twin model of production factors are constructed based on the analytic hierarchy process. For the digital twin model of the production system, first determine the characteristics of the digital twin production system model that require consistency verification, design requirements according to the characteristics to be verified, simulate, verify or predict based on the digital twin model of the production system, and compare and analyze the results with the actual measurement results of the physical experiment , to obtain the consistency verification results for the digital twin model of the production system. If the results are inconsistent, it is necessary to decompose and re-verify the assembled and fused model until the problem of the model is determined, and its parameters are corrected.

In order to further adjust and optimize the production process model, it is necessary to further realize model self-iteration on the basis of model self-perception. ①On the basis of constructing the mapping model of workshop elements-production line mapping state, based on the generative modeling method of autoencoder and generative confrontation network, a digital twin inverse model with workshop mapping state characteristics as input and workshop element data as output is constructed ;②Based on the inverse model, by changing the input of workshop state characteristics, observing and analyzing the changes of workshop element data, mining the influence law of workshop state characteristics on workshop element data and the coupling relationship between workshop state features, and realizing the mapping mechanism in the mapping model ③According to the real-time change law of the workshop state characteristics and its influence on the workshop elements, dynamically update and adjust the corresponding feature weights in the state mapping model, and at the same time adjust the feature extraction layer according to the coupling relationship between the workshop state features Structure and connection methods, so as to realize the dynamic self-iteration of the data-driven workshop model.

4.3.3 Virtual-real consistency verification of digital twin model

(1) In terms of workshop element models, virtual models of workshop equipment such as coating machines, ovens, and production lines for welding, liquid injection, and packaging are formed based on the assembly and integration of constructed geometric models, process data models, motion models, and rule constraint models. Model, each production line model is further assembled and fused to form a new energy vehicle power battery production digital twin workshop model. These models describe the physical workshop from multiple dimensions such as geometry, behavior, and rules, and multiple granularities such as unit level, system level, and complex system level.

(2) Dynamic modeling of production process Workshop data is an important driver to support workshop dynamic modeling. In order to realize the dynamic modeling of the production process, the problem of data source, that is, data acquisition, must be solved first. 1) In terms of data collection, workshop data has the characteristics of multi-source heterogeneity. The traditional workshop data collection method is manual collection, but this data collection method has low efficiency, low accuracy, and long lag time, which cannot meet the needs of automation and intelligence. need. Then came the use of barcode [23], RFID [24], sensors and other methods for data collection. These methods have improved the efficiency of data collection to a certain extent, but the collected data has the problems of information islands and heterogeneous data fusion. 2) In terms of production process data modeling, some scholars have carried out feature extraction at different time and space scales from the perspectives of time domain [25], frequency domain [26], and time-frequency domain [27,28] to explore the production status. The relationship between features and data and the establishment of a dynamic mapping model of the production line state, but the extracted features are numerous, there is redundancy, and the correlation between different features and states is quite different, so it is difficult to build an accurate production line process model. 3) In terms of iterative update of the model, the operation process of the workshop is complex and changeable, the data flow is large, and the real-time requirements are high, which makes it difficult to update the production process model in time. The above factors such as heterogeneous, multi-source and isolated data, and unclear mapping mechanism between data and production process make the production process model inaccurate/not timely in describing the dynamic production process, resulting in poor timeliness of production decision-making.

(3) Production system simulation modeling 1) In terms of setting simulation constraints, current simulations mostly focus on fixed constraints such as process constraints [29], resource constraints [30,31], performance constraints, and time constraints [32]. Considering the passive constraints that may be brought about by dynamic abnormal events, the simulation process model has insufficient adaptability to the dynamically changing production environment. 2) In terms of simulation methods, one type of method is based on the mechanism model for simulation [33], and the other type is based on the data model for simulation [34]. There is insufficient research on the simulation method based on the fusion of data and mechanism model [35] , it is necessary to further carry out research on the simulation process model of data and model fusion to overcome the shortcomings brought about by the single drive of the model or data. 3) In terms of simulation mode, most of the current simulations are aimed at a single purpose or function, such as workshop layout simulation [36], workshop scheduling simulation [37,38], workshop logistics simulation [39-41], global simulation of all-factor coordination Insufficient consideration leads to the one-sidedness of the simulation results. Inadequate consideration of the above simulation constraints, lack of in-depth integration of data and models, and single function of the simulation mode lead to a large gap between the simulation results and the actual process, which in turn makes the production and operation decisions inaccurate.

4.4   Dynamic modeling of production process driven by twin data

Aiming at how to mine data characteristics based on live data and solve the problems of twin data-driven production process state identification and dynamic iterative modeling, the solution is explained from three aspects based on production process multi-source data perception, production process dynamic model construction and production process model dynamic iteration Or process, as shown in Figure 5.

 4.4.1 Heterogeneous and multi-source data perception in the production process

The perception of multi-source data in the production process is the basis and premise for realizing the construction of a dynamic model of the production process. However, the types of equipment in the workshop production line are complex, the communication standards are diverse, the data sources are numerous, and the characteristics of multi-source heterogeneity lead to difficulties in data collection and management. Difficult to share. This paper builds a cloud/fog/edge collaborative data collection architecture to realize efficient perception and processing of production line information [43].

Different levels of cloud/fog/edge have certain differences in their data storage, processing, and analysis capabilities. Make full use of the capabilities of different levels to provide guarantees for multi-source data perception in the production process. (1) At the edge layer, for the personnel, equipment, materials, processes, environment and other data in the workshop, according to the data type, value and the transmission protocol followed, the collection is realized through equipment connection management, and the equipment with the same communication protocol is set Under the same data acquisition device, it is convenient for batch processing, and feature extraction is performed on high-dimensional data to reduce the amount of data transmission and increase the transmission speed, and then transmit them to the fog layer for post-processing. (2) In the fog layer, the data from the edge layer following different communication protocols are converted into a format conforming to the OPC UA unified architecture through the information model and integrated. The server can transmit the data to the cloud through the publish-subscribe method, or through The C/S mode realizes the communication with the client, and then realizes the storage or use of data. (3) In the cloud layer, store data from the fog layer, build a relational database, facilitate data management and call, and use the huge computing resources in the cloud to train the mapping model of the production line state. In addition, since the OPC UA unified organization supports the "write" function, combined with method calls, data and commands are processed. The cloud can not only obtain the collected data and information, but also directly send control and correction instructions to the edge devices to achieve optimal control of the devices, thus forming a complete closed loop of cloud-fog-edge.

 4.4.2 Construction of dynamic model of production process

In the process of product production, there are many factors that affect product quality, and the state of the production line is coupled with each other, so it is difficult to build a more accurate dynamic change mechanism model of the production line. Extracting data features, establishing the mapping relationship between production line data and production process state, and analyzing the mapping mechanism are the key points of dynamic modeling of production process in workshop digital twin.

Aiming at how to analyze the constraints and coupling relationships among elements, models, and data in the system, establish a system collaborative simulation model to realize the simulation analysis and decision-making of the production process, from the construction of production system simulation constraints, simulation model construction, and simulation model reduction3 A specific solution is described in each aspect, as shown in Figure 6.

The basic feature extraction method of data can extract features such as maximum value, minimum value, average value, variance, and skewness in the time domain, and can extract features such as average frequency, root mean square frequency, and frequency standard deviation in the frequency domain. For different application requirements, application scenarios, and application objects, there are also certain differences in the construction process and method of the dynamic model of the production process. (1) The health of the equipment in the production process is an important factor affecting the quality, production capacity and safety of the product. In the process of modeling the health state of production line equipment, the cutting force, current and other data of the equipment processing process can be collected, and the health factors of the equipment state can be established according to the self-organizing map [44] long-short-term memory network, and the mapping between the data and the equipment state can be realized. At the same time, the failure threshold is set to estimate the remaining life of the equipment, and then the equipment can be predictively maintained in time. (2) The production of products is the fundamental purpose of production line operation. In order to improve the efficiency of parts quality inspection in complex manufacturing environments, the characteristics of production line data can be screened according to correlation analysis, and the type 2 fuzzy neural network [45], Gaussian Algorithms such as process regression and principal component regression establish the mapping relationship between signal features and part quality, thereby building a model of part quality and realizing virtual measurement of part quality and acquisition of quality features. (3) In order to realize the evaluation and optimization of energy consumption, according to the function and processing mechanism of the equipment, based on algorithms such as analytic hierarchy process and machine learning, a digital twin model of workshop energy consumption can be established for energy consumption perception, simulation and optimization.

 4.4.3 Dynamic iteration of production process model

Reasonable analysis and construction of digital twin model association relationship is the key foundation and premise to realize model assembly and fusion. Multi-scale twin model assembly and fusion are carried out from two aspects of production elements (parts-components-equipment) and production systems (equipment-production line-workshop). According to the spatial dimension of production factors, analyze the vertical, parallel, tangential and other spatial relationships among component models. For the spatial dimension of the production system, analyze the spatial layout relationship of equipment and the connection relationship at the mechanical level. Through the analysis of spatial relationship, a multi-layer structure tree that can truly describe the relationship between production factors and production system is established, so as to effectively describe the multi-scale relationship from unit level-system level-complex system level.

The assembly and fusion of the full-element multi-scale digital twin model is to assemble and fuse unit-level models (such as equipment) into a system-level digital twin model (such as a production line) by adding spatial relationships, constraint relationships, etc., and the system-level model can be based on Requirements are further assembled and fused into a complex system-level model (such as a manufacturing plant). First, based on the spatial relationship obtained from the analysis, different models are assembled in the same software environment at the spatial level. Then add its internal associations to the assembled model, and make the model have knowledge in various fields by mapping the action timing relationship, process constraint relationship, energy flow, information flow, material flow, etc. into the model.

(1) Modeling of production factors 1) In terms of building multi-dimensional models of production factors, the current research or application of digital twin factor modeling mainly focuses on the construction of geometric models to support the monitoring of workshop status [7], including using 3D software to directly Modeling [8], modeling using instrumentation and equipment measurement methods [9,10], modeling using video or images [11,12] and other methods. However, the multi-dimensional description of the workshop's physics, behavior, and rules is insufficient. 2) In terms of building multi-spatial-temporal scale models of production factors, in terms of spatial dimensions, current modeling mostly focuses on single-level objects such as key components [13], equipment [14] or production lines [15], and lacks a “unit level-system Systematic research on model assembly and fusion from a multi-level perspective. In the time dimension, although there are studies on different stages such as the design stage [16-18], the manufacturing stage [19] and the operation and maintenance stage [20,21], the research on the whole life cycle model needs to be further deepened. 3) In terms of the consistency verification of the production factor model, currently most of them use the design of specific experiments to verify the correctness of the model [22], but the results of the designed experimental verification cannot fully reflect the accuracy of the model, so The current workshop element modeling requires a standardized and unified model consistency verification theory and method, so as to improve the accuracy of the model. To sum up, workshop element modeling lacks in-depth research on issues such as accurate construction of multi-dimensional models, assembly and fusion of multi-spatial-temporal scale models, and model consistency verification. Realize precise workshop control.

In order to realize the dynamic iteration of the production process model, it is first necessary to be able to perceive the dynamic changes of the production process, that is, to realize self-perception. ①Use methods such as model pruning, knowledge distillation, and reinforcement learning to compress and reconstruct the mapping of the production process state, reduce the complexity of the mapping without losing the accuracy of too many mapping relationships, and obtain the reconstructed mapping model; ②Through the reconstruction The online training of the mapping model obtains the real-time perception model of the workshop. On the one hand, the real-time perception model of the workshop retains part of the mapping ability of the state mapping model, and on the other hand, it includes the dynamic factors of the production line after the real-time data training of the production line; The state-aware model and the original state-mapping model are fused based on the model of structure intersection, so as to realize the dynamic self-awareness of the workshop state-mapping model driven by real-time data.

 4.5    Simulation Modeling of Production System Driven by Model and Data Fusion

 4.5.1 Construction of production system simulation constraints

Reasonable setting of production system simulation constraints is the foundation and premise of production system simulation modeling. This section constructs two types of constraints including actively set constraints and constraints caused by abnormal events from the two dimensions of the multidimensional model of production factors and the dynamic data of the production process.

Actively set constraints are based on the process characteristics, resource requirements, and operational performance of production factors, and consider actual production tasks and goals to construct constraints in terms of process constraints, resource constraints, performance constraints, and time constraints for the production system. The constraints caused by abnormal events are real-time constraints driven by the dynamic data of the production system. The construction process includes the construction of disturbance type feature library, feature information extraction and optimization, and disturbance identification. Firstly, around the dynamic and variable production environment, by analyzing the types of disturbance events and their representations, construct a model that includes explicit disturbances (such as urgent mail arrivals, equipment failures, delivery changes, material shortages, scrapped rework, personnel delays, etc.) and invisible disturbances (disturbances) Such as man-hour fluctuations, workpiece completion time deviations, changes in the number of buffer parts, changes in the number of warehousing parts, etc.) twin dynamic event type feature library of dynamic production process disturbances; use the dynamic data of the production process, based on the hidden Markov model Data is extracted, and the extracted results are streamlined and optimized through DS evidence theory to obtain optimized feature information; the convolutional neural network is used to build a correlation model between the disturbance representation in the disturbance type feature library and the disturbance information of the production line, and the convolutional neural network is trained to The network realizes the identification of recessive and explicit disturbances in the production line, obtains twin dynamic events, and constructs constraints caused by abnormal events.

 4.5.2 Construction and update of production system simulation model

The construction of the production system simulation model is based on the multi-dimensional multi-domain model of the workshop elements and the dynamic data of the production process. By configuring the workshop element model to form an application-oriented digital twin model of the production system, and then applying constraints, initialization and input settings to the simulation model based on the dynamic data of the production process, the accurate construction of the simulation model is realized.

The specific construction process of production system simulation model construction includes the following three steps: (1) Based on the digital twin model of workshop elements, guided by application demand analysis, according to the collaborative interaction between stations and equipment, the AHP is used to automatically The configuration forms a digital twin of the production system. (2) Add actively set constraints and constraints caused by abnormal events to form a production system simulation model. (3) Initialize the built production system simulation model based on the dynamic data of the production process, so that the initial state of the simulation model is consistent with the actual production. In addition, before the simulation solution starts, the simulation model is obtained by analyzing the dynamic data of the production process and setting the input of the simulation model. Due to the attenuation of production line performance, changes in working conditions, and adjustment of production tasks, the immutable model cannot meet the application based on the digital twin model of the production line. Therefore, it is necessary to quantify the degree of state change through perception data, and perform performance evaluation on the model based on hierarchical analysis and update rules. Status update. Finally, the updated production system element model is combined with task status to obtain a global update model.

 4.5.3 Order reduction and solution of production system simulation model

Due to the substantial increase in the income of migrant workers in recent years, the young and strong rural labor force generally believes that growing vegetables is not as good as working, and there is no guarantee for growing vegetables, so they go out to work more. The vegetable industry is an industry with high capital investment, high technical requirements, and high labor intensity. Most of the vegetable growers in rural areas are middle-aged and elderly people and women, and their ability to learn and apply new technologies is low, which affects the output and benefits of growing vegetables. In addition, the propaganda, training, and guidance of farmers are not thorough enough, the concept of change is slow, and the degree of participation is not high. The development of the vegetable industry still relies too much on the promotion of the government, and the main role of farmers has not been fully utilized.

The production system simulation model involves many and complex digital twin models, and the production process data that needs to be processed is diverse and huge. In order to meet the accurate and efficient real-time simulation solution requirements of the production system simulation process, the model needs to be reduced and optimized first. Based on the multi-agent distributed computing and distributed heterogeneous computing scheduling algorithm, the simulation solution is realized.

The order reduction of the production system simulation model in this paper refers to reducing the order of the equations in the full-factor simulation process, maximizing the solution efficiency while preserving the solution accuracy of the process simulation model, and then obtaining application-oriented services. High-fidelity lightweight production system process simulation model. Firstly, taking the application service of the production system as the orientation, the requirements and characteristics of the application service are analyzed, and the application service characteristics and requirements are used as the evaluation criteria, and the maximum error limit in the time domain is obtained by means of the AHP; then, the maximum error limit in the time domain is used The impact evaluation method and model reduction algorithm (Krylov subspace method, balance stage and other algorithms) are used to reduce the order of adaptive error limit; finally, the dynamic link library obtained by encapsulating it is used to self-optimize and reduce the order of the full-factor model of the production system processing, generating a simplified reduced-order model that preserves the necessary behavioral properties and dominant effects of the production system. For the reduced-order production system simulation model, by dividing large-scale and complex manufacturing resources into several agents, the task decomposition, assignment and collaborative solution of multi-agent distributed collaborative solution are carried out to realize complex problems in dynamic and changeable environments. Fast decision-making, and the use of distributed heterogeneous computing resource scheduling algorithms for intelligent production systems to achieve efficient real-time processing of production system simulation models, providing support for application services such as production process value analysis, production equipment failure prediction, and production cycle decision-making .

5 Case: Construction of digital twin workshop for new energy vehicle power battery production

The power battery is a key component of new energy vehicles. Its manufacturing process is complex, and the control of each process in the production process will have an impact on the quality of the battery. In order to realize the precise management and control of the power battery production process, the author's team Machinery Industry Sixth Design and Research Institute Co., Ltd. combined a new energy vehicle power battery production workshop with a new project and applied a new generation of information technology to build a new energy vehicle power battery production workshop The digital twin model, as shown in Figure 7.

The new energy vehicle power battery physical production workshop includes key equipment such as batching tanks, coating machines, tunnel furnaces, roller presses, slitting machines, die-cutting machines, stacking machines, elevators, and packaging machines. The batching, coating, rolling, slitting, die-cutting, lamination, welding packaging and other processes of production. The construction of the digital twin workshop for new energy vehicle power battery production includes the following three aspects:

• By clicking the "Add Student Information" button in the main interface, enter the inputDialog interface shown in Figure 2. The InputDialog interface class is automatically generated by adding and inheriting from the QDialog class under the interface file in the Qt Creator development environment. In order to realize event processing and interface jump in Qt, you can add a slot function to the signal function of the control. When the "Add Student Information" button in Figure 1 is clicked, a click signal is sent, and then on_inputButton_clicked is added to the click signal function. () slot function to realize the jump from the main interface to the interface of adding student information.

• In terms of the production process model, workshop twin data includes state data, switch data, video images, etc. collected from the physical workshop, as well as workshop SCADA and MES system data, such as order information, production plan, actual output data, quality inspection data, etc. Based on the processing and fusion of these data, the multi-dimensional analysis and optimization of the production process status can be realized.

• In terms of the simulation model of the production system, based on the virtual model and twin data, the simulation models of the positive electrode coating process and the negative electrode coating process were constructed. The relevant models provide important support for the visual monitoring of the production process in the workshop, the simulation of the coating process, and the production planning from raw materials to finished products.

6 Conclusion

The construction of the digital twin model is an important prerequisite for the realization of the digital twin function. In order to make the digital twin model construction process evidence-based, this paper proposes a set of digital twin model "four modernizations, four uses and eight uses" construction criteria, and based on the proposed digital twin model construction criteria, explores and builds a set of "construction —Group-Integration-Verification-Calibration-Management" digital twin model constructs a theoretical system, and conducts practical exploration in the digital twin workshop. The related work is expected to play a reference role in the research of digital twin model and its application in the workshop.

This article is only for learning and communication, if there is any infringement, please contact to delete! !

Add V "Artificial Intelligence Technology and Consulting" for more information! !