In real manufacturing industry, the timely identification of weaknesses and precise determination of optimization areas are crucial for efficient workshop production. Traditional plant simulation methods have been widely employed to support decision-making, but the accuracy of these decisions heavily depends on the reliability of the simulation results. Static simulations, which rely on manually set parameters often fall short of meeting these demands. To address this, dynamic simulations incorporating real-time data are essential for generating accurate and actionable insights. A worth-trying approach is proposed therefore in this paper by establishing a plant simulation environment in Digital Twin(DT) utilizing a high-fidelity real-time coevolution model for intelligent manufacturing cells. To get the coevolutionary models in DT system, data modeling and integration, virtual-real event driven method are illustrated. This approach provides a true-to-life workshop simulation environment. By employing variable and customizable optimization strategies, diverse simulation outcomes can be generated, aiding in the selection of the most effective optimization strategies. The development of a coevolutionary DT for a manufacturing cell within an enterprise is presented as a case study, demonstrating the effectiveness of dynamic and combo-strategy production simulations.
Research Article
Approach of Dynamic Simulation and Optimization for Coevolutionary Intelligent Manufacturing Cell in Digital Twin
https://doi.org/10.21203/rs.3.rs-4965442/v1
This work is licensed under a CC BY 4.0 License
You are reading this latest preprint version
manufacturing cell
coevolutionary model
dynamic simulation
digital twin
In today’s competitive landscape, flexible manufacturing cells are increasingly adopted by enterprises to boost production efficiency and achieve economic benefits. Given the inherent uncertainty and complexity of manufacturing environments, digital simulation stands out as the most effective method for early prediction, warning, and optimization. The benefts of digital simulation go hand in hand with the main targets of manufacturing organizations such as reducing cost and waste while simultaneously enhancing quality, efficiency and safety. It depends on real dynamic and exact digital environment. Fortunately, the evolution of advanced manufacturing systems is steering towards digital factories within Cyber-Physical Systems (CPS), which integrate mechanical, electrical, software, and communication components, all interconnected with their corresponding physical plant environments. This progression is making digital dynamic simulation increasingly feasible and practical for modern manufacturing.
The Digital twin (DT) is a sophisticated technical approach that aims to create a digital mirror of a physical system. Within this framework, CPS serves as virtual models in the DT, enabling continuous data exchange between the physical and digital realms. Leveraging DT's real-time mapping capabilities, integrating diverse simulation strategies emerges as a novel solution for dynamic simulation. This integration can not only enable ongoing digital monitoring, but also identify production weaknesses and active functional improvement or decision optimization by establishing a high-fidelity real-time mapping model of the manufacturing cell. Additionally, this represents another valuable application of DT, which continues to be a focal point for researchers and managers in the enterprise sector.
The concept of the Digital Twin was first introduced by Grieves in 2003. It serves as a digital representation of physical entities, capable of dynamically reflecting the current state and behavior of physical systems in real-time, as well as predicting their future states and behaviors[1–2]. In 2017, Tao Fei et al. explored the application of Digital Twin technology in workshop environments [3–5], categorizing DT into two distinct stages: physical data integration and cyber-physical data integration. The first involves collecting massive data from different manufacturing assets in physical world and transmitting those data into the DT systems. However the second one is the essential part of DT, which signifies to add value and efficiency of both the data collected from the physical manufacturing assets and the data generated or existing in these deployed digital systems.
Arround these two stages, recently versatile researches have been done on DT workshop in different scenario. In 2019 Tao Fei[6] proposed a DT workshop frame with five-dimensional model which described the architecture how to integrate the physical and virtual world. To address the challenge of data mapping in virtual spaces, Zhou Cheng et al.[7] proposed a three-dimensional visual monitoring system under the DT conception by industrial IoT platform collecting all data from an seal production workshop. While K. Vijayakumar et al.[8] suggested employing Iot for real time locating and sensing of all factory facilities. No matter the facilities were pluged in or not, moving or keeping steady, it helped to keep all the the models under surveillance and simulate in real time. To coordinate human workers and autonomous systems, sharing the same assembly environment of the interior of an aircraft fuselage, Franceschi P. et al.[9] developed a prototype that mirrored the plant system, displaying its status on a dashboard and equipping human operators with essential tools for supervision and rapid intervention in case of failure. These applications highlight the significant attention given to online physical data integration within DT systems.
Recent literature reports indicate that researchers are delving into more valuable and advanced applications of DT technology, leveraging specialized models within the DT. Ye Lei et al.[10] proposed a DT-based pre-spinning workshop in textile factory. It not only assisted operators quickly locate the accident spot by monitoring the state of spinning machine, but also optimized the working path of the robots with some prediction models in DT. For complex products like satellites, missiles, and aircraft typically have demands for dynamic data management and process traceability, Zhuang et al.[11–12] implemented a DT solution involving one-to-one mapping of the corresponding physical entities, and fulfiled the synchronous modeling of the product assembly process, recorded the real process data for each product. Additionally, Liu et al.[13] combined DT with scheduling algorithms to develop an intelligent workshop scheduling cloud platform.
Furthermore, simulation based on DT has garnered significant interest. Yang et al.[14] constructed a DT system that mirrors mirrors the actual spacecraft assembly process, allowing real-time reflection of assembly issues and enabling corresponding behavior simulations. By comparison of simulation and anysis, DT may provid a feasible solution. Similarly, Söderäng et al.[15] explored the feasibility of a fully physics-based combustion engine model in real-time co-simulation with an electrical power plant model and developed a DT engine model in Simulink. Using the measurement data from the infrastructure provide validation for the digital twin, the model-in-the-loop simulation results proved the accuracy when compared to the measured values.
From all above, they are valuable references, especially presented those collection and integration methods for manufacturing data from physical assets, proposed the applications in particular optimization for manufacturing systems in different industries. However, one of the significant challenges remains in determining what should be optimized and which models to use, especially given the inherent complexity and flexibility of modern workshops. Without a clear understanding of the real-time state within the workshop, it becomes difficult to identify optimization opportunities accurately. Therefore, an effective production simulation system is essential to pinpoint precisely where and what needs to be optimized. Many of the existing studies on production simulation in workshops are based on predefined input conditions and parameters, resulting in offline simulations with non-real-time data[16–18]. And some references focus solely on dynamic simulation models for production planning[19–22]. But it is hard to satisfy the demand of real industry. Enterprise managers need to foresee the future efficiency and identify whether the production can be guaranteed. If not, they need to understand whether there is still an opportunity for optimization and improvement.
In response to these challenges and based on a current project in implementation, we have developed a DT system for an intelligent manufacturing cell, tailored to the real-world needs and key concerns of enterprise managers. This DT system, focused on production optimization, is driven by real-time data from the physical world and enables simulations based on coevolved cell models. The main contribution point is that the DT system serves not only as a virtual mirror of the manufacturing cell but also as a platform for optimizing and executing strategies through simulations based on real-time data from the physical cell. It highly improves reliability and accuracy of the simulation result, as well as possesses the flexible simulation strategy to deal with variable situations.
As a core environment for dynamic simulation in DT solutions, the digital shop floor serves as implementation and operation framework for Digital Twins (DT), which requires a broad horizontal and vertical integration [23], e.g. including consolidating, analyzing, connecting data derived from sensors and actuators, processes, production machines. With reference to the five-dimensional model of the DT [24], the architecture for such horizontal and vertical integration system is built which is shown in Fig. 1.
The physical cell layer serves as the foundational component of the DT system. The assets within the physical cell are categorized into various groups, including software (e.g., ERP, MES), sensors (e.g., RFID, acoustic emission, vibration, thermocouples), machine tools, and logistics equipment (e.g., robotic arms, AGVs, pallets, storage buffers). For machine tools, the data from CNC system is collected by NC-Link interconnection protocol. The data from sensors is get through the programme for acquisition card. The logistics is under control by PLC system, and its data is collected based on OPC. Data transmission module is deployed based on MQTT and socket in digital twin environment in order to subscribe the high frequently varied real-time data which is published to MQTT server. And for those static data from software (like the task schedule, the storage in central tool magazines, the occupation state of pallets), can be get by running certain APIs.
All data collected from the physical cell is processed using the JSCode (JsonCode) event method, which drives the synchronous evolution of the twin model through behavior and rule models, as illustrated in the 'Virtual Cell' section of Fig. 1. And in the top service layer, it provides real-time monitoring and prediction for safety, dynamic simulation for production, data statistical analysis, and strategy optimization.
Obviously data acquisition is not sufficient according to the function of the DT. It is necessary to achieve a virtual environment that can fully reflect the real state of the physical workshop, and at the same time, optimize according to the actual execution situation, and feedback the optimized rules to the control system. We name this as coevolution of the manufacturing system under digital twin. To realize these functionalities, both virtual-physical mapping and real-time simulation and optimization are essential. To achieve effective mapping between the digital and physical domains, we employed a data-event-action-based behavior model, addressing the challenge of virtual synchronization. This behavior model accurately depicts the real-time responses and behaviors of physical entities, considering both external environmental factors and internal operational mechanisms.
The behavior model serves as the cornerstone for the coevolution of the DT manufacturing cell, ensuring seamless synchronization between the DT system and its corresponding physical counterpart. Actually the behavior model rely on not only the realtime data but also the precise virtual-physical mapping. Thereby the behavior model, Data-Event-Action, was embed inside DT cell.
3.1 Event description by JSONCode
The collected data like manufacturing status and the interactive information among various physical assets will be parsed into corresponding events, which is the process of Data-Event. In order to accurately describe the manufacturing activities that occur in the physical space and facilitate the virtual space to validate process data, a unified data and events description format based on JSONCode has been defined in the digital twin data layer. It mainly consists of the following fields as shown in Fig. 2.
In Fig. 2, Flag indicates the success status of data requests and transmissions, while CodeData represents a set of data information, CodeID and CodeName is the identifiers for the data information, Priority denotes the priority level assigned to the data, StartTime and Endtime represents the time for execution of information, Parameters refer to the data set of information, DeviceState indicates the state of the device, CurDev represents the current device where the material is, CurPos denotes the position of the material on the device, TarDev is the target place for the material, and Message refers to the message returned by the server. The set of Parameters can be dynamically adjusted based on the specific parameters of each device.
3.2 Embeded logics and rulls trigged by events
The collected data is structured into JSONCode, which, when transmitted to the Digital Twin environment, is parsed by specific programs into event information corresponding to various entities within the system. Such event information will be assigned to the corresponding assets inside DT. It acts like a trigger switch, activating the behavior processes of assets according to some pre-established logic and rull programs. It is the process of Event-Action, which ensures the DT system acting like a virtual mirror of the physical workshop.
Figute 3 The sequence diagram for assets
Those logic and rule drive the behavior of virtual assets in DT to be consistent with the behavior of actual manufacturing cell, just like a mirror. During the actual production process, the operation logic is concerned with multiple objects. For instance, in the process of loading a cutting tool from the central tool magazine to the machine tool magazine, the cutting tool, tool robot, and both magazines are components involved. Therefore, the logical process can be decomposed into separate actions. When the tool robot receives the command and arrives at the specified location of the central tool magazine, it will fetch the cutting tool and move toward the specified machine tool after checking the status of the central tool magazine. Confirming the position ID and available state of the tool magazine inside the machine, the robot will execute the tool loading. Meanwhile, the state of the two magazines will be updated, the successful execution information will be sended back to control system, and the robot will return to home position. Similarly, those complex operations can be decomposed as well based on object-oriented as shown in Fig. 3. According to this sequence diagram, the logical behaviors of each asset can be programmed.
Leveraging the Data-Event-Action model, the Digital Twin's virtual mirror can be effectively realized. However, the DT system for a manufacturing cell should transcend mere behavior replication, functioning also as a strategic simulator. The purpose of simulating manufacturing cell is to gain an understanding of the current state of the units and to further analyze and predict whether the subsequent production processes will meet the production schedule on time. What is the production capacity of the cell? What will be the asset utilization rate in the future? Do different execution rules have an impact on the production process, and so on? The above-mentioned questions can be addressed with the assistance of dynamic simulation analysis and forcasting based on real-time cell status.
The primary limitation of traditional simulations is their lack of real-time capability, which hinders their ability to promptly simulate and analyze future performance indicators of the manufacturing cell based on current production data. Yet the ability to optimize these indicators in the future is a critical issue of concern to managers. The limited data integration between traditional simulation systems and manufacturing execution control systems often results in simulations that rely on historical or artificially generated data. This phenomenon significantly reduces the credibility of simulation results. To address this issue, this paper proposes the use of actual real-time data based on digital twin environment, geared towards the future production load of manufacturing cell, to conduct genuine dynamic simulation.
4.1Data Integration during Dynamic Simulation Time Zone
Performance analysis of the manufacturing cell largely depends on the statistical evaluation of process data. However, the DT system is not the actual execution control system of the cell. Instead, the Manufacturing Execution System(MES) primarily focuses on the control of the production process and often does not need to pay attention to numerous tedious asset assessment data, such as the actual mileage of transport vehical, the frequency of tool changes in the central tool magazine, the cumulative waiting time of trays in the pallet warehouse and so on. These types of data can be easily obtained in a twin environment. Therefore, before conducting simulations, it is necessary to use the required simulation window time to integrate production tasks, manufacturing process data from the unit manufacturing execution system with the assets assessment information recorded in the DT, forming the input foundational data source and initializing the dynamic simulation environment, as shown in Fig. 4
4.2Simulation Strategies for Different Objects
After configuring the initial environment, the simulate execution involves a loop consisting of two steps. The first step is to analyze cell performance to identify areas of dissatisfaction and potential improvements. The second step is to adjust strategies based on the assessment of cell performance, conducting simulations to provide forecasts for the cell performance.
The key aspect here is the adjustment of strategies, as the manufacturing process can be influenced by various unforeseen circumstances, such as equipment failures, changes in production schedules, and revisions to manufacturing processes. Additionally, the goals that the cell needs to achieve may also change at different stages. For example, approaching the delivery deadline may require the primary goal to be timely completion, while in a stable process with batch production, the goal may shift towards improving production efficiency and reducing manufacturing costs. Therefore, in response to such changes and to achieve specific goals, it is necessary to adjust the strategies of the cell’s resources.
The overall operation logic of the cell can consist of the operation strategy of multiple asset objects. When adjusting the operation strategy of the cell, on the basis of the formulated strategy for different objects, the production and manufacturing process of the cell can be optimized through the combination of strategies, so as to improve the performance of the cell and achieve different goals. Table 1 outlines distinct operational strategies tailored to various production conditions, each specifically designed to optimize the performance of different components within the manufacturing cell.
| Strategy ID | Strategy name | Description |
|---|---|---|
| ① | Pallet Storage Position Locked | The storage location of the pallet is bound to the pallet ID and cannot be stored randomly. |
| ② | Pallet Stores Sequentially | Pallet will be stored according to open spot sequentially. |
| ③ | Pallet Stores in Closest Open Spot | When an AGV transports a pallet to the storage, it will select the closest open spot based on current AGV’s own position. |
| ④ | 5-axis Machine Tool Can Process 4-axis Parts | When the idle time of the 5-axis machine tool exceeds more than half of its working time, while the tasks on the 4-axis machine tool are overloaded, in order to ensure production schedule and balance machine utilization, the five-axis machine can be assigned the task of machining four-axis parts. |
| ⑤ | 5-axis Machine Tool Can NOT Process 4-axis Parts | Five-axis machine tools can only process five-axis parts even it is idle. |
| ⑥ | Machining task can not switch to the replaced machine tools. | When the machining task is assigned to certain machine tool, it can not be switched to another machine tool even it is the same type. |
| ⑦ | Machining task can switch to some replaced machine tool | When the machining task is assigned to certain machine tool, it can be switched to another machine tool even it is the same type. |
| ⑧ | Workpiece unload rull with operator on duty | When operators are on duty, the machined workpiece is prioritized and sent to the unloading position instead of the pallet storage. |
| ⑨ | Workpiece unload rull without operator on duty | When the cell is unmanned, the loading and unloading station is locked and the machined workpiece should be sent to the pallet storage. |
| ⑩ | Shortest path for AGV | AGVs will execute optimization algorithms based on the pending destinations in the tasklist, thereby reducing redundant paths and obtaining the shortest route. |
| ⑪ | AGV strictly follows the tasklist | The AGV strictly follows the order of destinations and cannot skip any destination. |
| … | … | … |
| For the simulation strategies in the above table, mathematical modeling is carried out separately, and the algorithm formulas are shown below by setting \(\:{Str}_{i}\) as the strategy: | ||
Strategy ①-③ means that the AGV should store the pallet in certain position according to the strategy option. The formula is:
$$\:AGV\_PalletDestination（TaskID）=\left\{\begin{array}{c}Pallet(Allocated\_StoreID)\begin{array}{cc}&\:\begin{array}{cc}&\:\end{array}\begin{array}{cc}&\:\end{array}\begin{array}{ccc}&\:&\:\end{array}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:{\:\:\:\:\:\:\:\:Str}_{1}=1\end{array}\\\:Pallet\left\{Min\left(StoreID\right)\right\},StoreID\in\:\left({State(StoreID}_{i}\right)=\text{I}\text{S}\text{O}\text{P}\text{E}\text{N})\begin{array}{cc}&\:\:\:\:\:\:\:\:\:{Str}_{2}=1\end{array}\\\:Pallet\left\{MinLength\left(StoreID\right)\right\},StoreID\in\:\left({State(StoreID}_{i}\right)=\text{I}\text{S}\text{O}\text{P}\text{E}\text{N})\:{\:\:Str}_{3}=1\end{array}\right.$$
1
Here, \(\:MinLength\left(StoreID\right)\) is the closest spot to store the pallet which\(\:\:\)can be calculated with the formula:
$$\:MinLength\left(StoreID\right)=Min[PathFind\left({AGV}_{pos},{\:StoreID}_{i}\right]$$
2
And where \(\:{AGV}_{pos}\) is the AGV current position and \(\:i=\text{1,2}，\dots\:\:\text{M}\text{a}\text{x}\text{i}\text{m}\text{u}\text{m}\:\text{s}\text{t}\text{o}\text{r}\text{a}\text{g}\text{e}\).
Strategy ④-⑦ represents whether or how the machine tool can be replaced. The formula is shown as following:
$$\:\left\{\begin{array}{c}\begin{array}{cc}{Mac}_{4axis}\left({TaskList}_{j}\right)\underrightarrow{switch}\:{Mac}_{5axis}\left({TaskList}_{k}\right)&\:{Str}_{3}=4\:and\:{RTO}_{IdelTime}\left({Mac}_{5axis}\right)\ge\:50\%\end{array}\\\:\begin{array}{cc}{Mac}_{4axis}\left({TaskList}_{j}\right)\underrightarrow{No\:switch}{Mac}_{5axis}\left({TaskList}_{k}\right)&\:{Str}_{3}=5\:and\:{RTO}_{IdelTime}\left({Mac}_{5axis}\right)\le\:50\%\end{array}\\\:\begin{array}{c}\begin{array}{cc}{Mac}_{iaxis}\left({TaskList}_{j}\right)\underrightarrow{switch}\:{Mac}_{iaxis}\left({TaskList}_{k}\right)&\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:{Str}_{3}=6\end{array}\\\:\begin{array}{cc}{Mac}_{iaxis}\left({TaskList}_{j}\right)\underrightarrow{\:No\:\:switch}\:{Mac}_{iaxis}\left({TaskList}_{k}\right)&\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:{\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:Str}_{3}=7\end{array}\end{array}\end{array}\right.$$
3
Where \(\:{Mac}_{4axis}\left({TaskList}_{j}\right)\:\) represents Task j in 4-axis machine tool, and \(\:{RTO}_{IdelTime}\left({Mac}_{5axis}\right)\) means the idle time ratio of 5-axis machine tool.
Strategies ⑧-⑨ dictate that when an operator is online, machined parts are prioritized for transfer to the loading and unloading station. Conversely, when the station is unmanned, it is locked, and parts are directed to storage, awaiting shipment to the next process. The condition formula is:
$$\:\left\{\begin{array}{c}{Machine{d}_{Workpiece}}_{i}\left(Destination\right)={StoreID}_{i}\:\:{State(StoreID}_{i})=ISOPEN\:and\:OperatorISOFF\\\:{Machine{d}_{Workpiece}}_{i}\left(Destination\right)=LU\_Station\:\:State\left(L{U}_{Station}\right)=ISOPEN\:and\:OperatorISON\end{array}\right.$$
Where \(\:{Machine{d}_{Workpiece}}_{i}\left(Destination\right)\) represents the destination place for the machined \(\:{wokpiece}_{i}\), and the boolean constants \(\:\text{O}\text{p}\text{e}\text{r}\text{a}\text{t}\text{o}\text{r}\text{I}\text{S}\text{O}\text{F}\text{F}\) and \(\:\text{O}\text{p}\text{e}\text{r}\text{a}\text{t}\text{o}\text{r}\text{I}\text{S}\text{O}\text{N}\) show that the operator is online or not, \(\:LU\_Station\) is the loading and unloading station.
Strategy ⑩ involves optimizing the AGV's task list based on destination points, utilizing a genetic algorithm to determine the shortest transportation path within the model:
$$\:\left\{\begin{array}{c}\text{m}\text{i}\text{n}\sum\:_{i=1}^{nTask}length\left({d}_{i}\right)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\\\:length\left({d}_{i}\right)=Distance\left({Task}_{i+1}\right)-Distance\left({Task}_{i}\right)\end{array}\right.\:\:\:\:$$
4
Where \(\:{Task}_{i}\:is\:the\:set\:of\:AG{V}^{{\prime\:}}s\:task\:list\).
Strategy ⑪ limits the adjustment of AGV tasks and following the principle of "first in, first out", which means the pending tasks cannot be adjusted.
It is obvious that the above strategies can also be expanded according to actual needs
4.3Combination of Multi-Object Simulation Strategies
In manufacturing cell composed of numerous manufacturing resources, it is often necessary to optimize the execution strategies of multiple resource objects and obtain appropriate combination strategies based on current production progress and assets status. This approach allows for the combined optimization of the cell's execution efficiency, enabling it to adapt to the current manufacturing conditions. This capability represents one of the most significant advantages of utilizing a Digital Twin in manufacturing cells.
Here the strategies are selected and combined based on the principles of mutual exclusion and independence, as shown in Fig. 5.
There are four strategy modules for storage, machine tool, operator and AGV. The strategies in same colour belone to same group. These strategies are independent of each other in different group, while the strategies within the group are mutually exclusive. It means in each group, just one strategy can be selected. This approach allows for flexible addition of policies within the group, flexible addition of policy groups, and even the addition of policy modules. Ensure that the simulation system can be expanded in a timely manner according to actual needs
Therefore, before executing the simulation, it is necessary to select strategies for each independent strategy module. For example, selecting strategies ①+④+⑦+⑨+⑩, and conducting simulations based on the current manufacturing situation, observing the impact of the selected strategy on the production of the cell. And based on the simulation results, it is possible to realiz the production status in future which may help to make appropriate decisions.
As part of a project focused on developing an intelligent manufacturing cell for high-precision hydraulic servo components (HPHSP-IMC), a comprehensive Digital Twin system was developed. This system was created using advanced tools such as Siemens Plant Simulation, SolidWorks, UG, 3DMAX, and JDK1.8, enabling detailed 3D modeling and dynamic simulation of the manufacturing cell. Figure 6 shows the HPHSP-IMC DT system, which is composed with 4 CNC horizontal machining centers, AGV system, central tool magazine, pallet stores, load and unload station. The 3D lightweight models were loaded into Plant Simulation, and their behavior models were coded.
HPHSP-IMC DT system not only follows and displays the realtime physical status, but also it has the functions of statistic analysis, prediction for anomaly, and optimization by realtime simulation.
5.1 Statistic for analysis and Monitoring for prediction
Data collection programme was applied, convering the information into JSONCode, which drives the real time motion of all the models in DT environment. The data transmited to DT system includes the machine tool states, orders information, control commands, processes requirements and NC codes, etc., which will be stored in realtime database.
When the physical cell issues the control orders, the commands will be pushed to EMQ server through MQTT client. The HPHSP-IMC DT system obtains these commands by subscribing to topics in the EMQ server (shown in Table 2) and converts it into JONCode events, which can be sent back to the virtual space through sockets (shown in Fig. 7).
|
Topics |
Function |
Issueed by |
Subscribed by |
|---|---|---|---|
|
ProcessTask |
Announce the orders to be machined |
Cell controller |
DT System |
|
MessageCenterEcodeState |
Command for action of all assets |
Cell controller |
DT System |
|
MacStateChannel |
Declare the state of CNC Centers |
Cell controller |
DT System |
|
Error Message |
System error information |
Cell controller |
DT System |
|
A1 |
NC Data from CNC Center A1 |
Data acquisition host |
DT System |
|
A2 |
NC Data from CNC Center A2 |
Data acquisition host |
DT System |
|
…… |
…… |
…… |
…… |
After DT system get information in JSONCode format, it may parse the JSONCode into certain events and drive the 3D models to reacting, following the behavior of physical world. In the DT environment, AGV moves materials in and out of warehouse, central cutter manipulator changes tools, machine tool processing and workers move materials, as shown in Fig. 8. At the same time, the Dt system can be displayed in real time on the dashboard of the workshop as well as the manager's desktop.
Some of the real-time data of the processing collected by the DT system can be directly displayed, such as the current status of the CNC machine tool. Some of the attribute indicators can be obtained after statistics and calculation, such as the completion rate of the production task according to the plan, the utilization rate of the machine tool, the AGV loaded running length, the vacancy ratio of the pallet storage, etc. And through the embedded models, tool collision and machine tool failure monitoring can be carried out, as shown in Fig. 9.
5.2 Optimization by Simulation in HPHSP-IMC DT
According to the dynamic simulation mentioned above, the process is shown in Fig. 10.
Once the simulation time window—comprising the start and end times—has been determined, the next step is to retrieve both historical and current status datasets of the DT cell from the control platform. These datasets serve as the foundation for initializing the virtual simulation environment. Then, applying the selected strategie combo to run the simulation. Check the prediction of the simulation results and compare to the expectations, if the strategies meet expectations, then apply the strategies to real manufacturing cell, if not satisfied, revise the strategies and run the simulation again. The simulation interface is shown in Fig. 11.
The parts need to be produced in HPHSP-IMC in the following period is shown in Table 3. The operation and task information is shown in Table 4 The lead time of these parts is in three weeks.
Table 3 Parts list
|
Part ID |
Time Consume(per each) |
Quantity |
Machine tool |
|---|---|---|---|
|
4FL-10/0–01 Housing |
125min |
140 |
4-axis |
|
4FL-10/0–015 Cap |
55min |
380 |
4-axis |
|
KZ16-2/0–01 Housing |
205min |
100 |
5-axis |
|
4BB-23/0–11 Oil Cap |
160min |
95 |
4-axis |
|
4K17-21A/02-101 Connector Box |
175min |
65 |
4-axis |
|
4BB-12/0-606 Back Housing |
110min |
45 |
4-axis |
The strategy combo used in real world is "5-axis Machine Tool Can NOT Process 4-axis Parts "+"Machining task can not switch to the replaced machine tools"+"Pallet Storage Position Locked"+"AGV strictly follows the tasklist" and the workhour is 24-hour a day. In HPHSP-IMC DT system, according the history information and the current part plan lefted to do, the simulation predicts that the 4-axis parts can not be accomplished in time while the 5-axis part(KZ16-2/0–01 Housing) can be completed almost a week in advance(shown in Fig. 12). And the availability of the run time for each machine tool, the AGV travel length are also shown in Fig. 12.
The initial prediction did not meet expectations. Consequently, the combo strategy was revised to include: "5-axis Machine Tool Can Process 4-axis Parts" + "Machining task can switch to some replaced machine tool" + "Pallet Stores in Closest Open Spot" + "AGV strictly follows the tasklist", with the system operating 24 hours a day. And run the simulation again. As shown in Fig. 13.
As illustrated by the simulation results in Fig. 12, it is evident that the task load for 4-axis machine tools is significantly higher compared to that of 5-axis machine tools. Under the the current strategy, the availability of run time of A3 and A4 will be much lower than that of machine tools A1 and A2, even it is declined to 0. However, after optimizing the combo strategy, the runtime distribution across the machine tools became more balanced and favorable compared to the previous setup. Also the AGV loaded running length ratio is increased. It is clear that the performance will be improved by applying the optimized strategy. So this dynamic simulation based on real-time data helps to making decisions. And the current execution strategy can be adjusted by down load the optimized combo strategy to real world through the integrated control system.
(1) The virtual digital twin system for intelligent manufacturing cell can be effectively established using Siemens Plant Simulation. Integrated with the data acquisition and execution control system, the virtual DT system not only reflects real-time operations but also coevolves with the physical world, guided by embedded behavior models and strategic rules. It is an important guarantee for reliability of dynamic simulation.
(2) Within this coevolution DT environment, the performance of manufacturing cell can be observed and predicted through simulation. The outcomes of these simulations provide critical insights into potential optimization areas, enabling proactive adjustments to execution strategies. This leads to optimized equipment utilization and production capacity, ultimately making the manufacturing cell more flexible and intelligent.
a. This paper is supported by the project financed by the State Administration for Science, Technology and Industry for National Defense(CN), and is funded by the National Natural Science Foundation of China(Grant: 51205201).
b. There are no competing interests.
c. Authors’s contributions:
- Sheng Leng mainly completes the conceptualization of coevolution for digital twin, methodology, simulation system design, strategy optimization, and writing the original manuscript.
- Fuxing Chen mainly completes investigation of the real manufacturing cell, modelling and programming for the dynamic simulation.
- Yunpeng Wei mainly forcuses on the performance analyzing, paper review and editing.
- Xianggui Yang mainly forcuses on real systeml demand analyzing and providing optimization solutions.
- Cong Wang mainly completes the system deployment, validation, and operation debugging.
- Grieves MW (2005) Product lifecycle management: the new paradigm for enterprises[J]. Int J Prod Dev 2(1/2):71
- Grieves M (2011) Virtually Perfect: Driving Innovative and Lean Products through Product Lifecycle Management[M]
- Fei TAO, Meng ZHANG, Jianfeng CHENG, Qinglin Q (2017) Digital Twin Workshop: A New Paradigm for Future Workshop[J]. Comput Integr Manuf Syst 23(01):1–9
- Ying, Cheng (2018) Yongping Zhang,Ping Ji,Wenjun Xu,Zude Zhou,Fei Tao. Cyber-physical integration for moving digital factories forward towards smart manufacturing: a survey[J]. Int J Adv Manuf Technol 97:1–4
- Tao F, Liu W, Liu J et al (2018) Digital twin and its potential application exploration[J]. Comput Integr Manuf Syst, 24(1)
- Tao Fei LW, Ran Z, Meng et al (2019) The five-dimensional model of the digital twin of and ten domain applications[J]. Comput Integr Manuf Syst 25(1):1–18
- Zhou Cheng SK, Ting L, Jiang Y, Cungui Juri Kingland. 3D visualization and monitoring system for workshop based on digital twin [J/OL]. Comput Integr Manuf Syst :1–18 [2022-03-31].
- Vijayakumar C, Dhanasekaran R, Pugazhenthi S, Sivaganesan (2019) Digital Twin for Factory System Simulation[J]. Int J Recent Technol Eng (IJRTE), 8(1s2)
- Franceschi P Mutti Stefano,Ottogalli Kiara,Rosquete Daniel,Borro Diego,Pedrocchi Nicola. A framework for cyber-physical production system management and digital twin feedback monitoring for fast failure recovery[J]. Int J Comput Integr Manuf ,2022,35(6).
- Ye Lei X, Hui W, Yong Z, Wenqi W, Xing A digital twin-based architecture for pre-spinning shop operations[J]. Manuf Autom 2021, 43(06):130–133
- Cunbo Zhuang,Jianhua Liu,Hui Xiong Digital twin-based smart production management and control framework for the complex product assembly shop-floor[J]. Int J Adv Manuf Technol 2018, 96(1): 1149–1163
- Zhuang C, Gong J, Liu J (2020) Digital twin-based assembly data management and process traceability for complex products[J]. J Manuf Syst 58(3):118–131
- Liu ZF, Chen W, Yang CB et al (2019) Intelligent manufacturing workshop dispatching cloud platform based on digital twins [J]. Comput Integr Manuf Syst 25(06):1444–1453
- Yang Wenqiang,Zheng Yu,Li Shaoyang Digital twin of spacecraft assembly cell and case study[J]. Int J Comput Integr Manuf 2022, 35(3)
- Söderäng E Hautala Saana,Mikulski Maciej,Storm Xiaoguo,Niemi Seppo. Development of a digital twin for real-time simulation of a combustion engine-based power plant with battery storage and grid coupling[J]. Energy Conversion and Management,2022,266.
- Liu M, Xie J, Wu H, Fu J, Ding G (2023) Research on Workshop Logic Modeling and Simulation Based on Finite State Machine [J]. J Syst Simul 35(04):853–861
- ChenYang Cheng SF, Li CL, Lee R, Jientrakul C, Yuangyai (2022) A Comparative Study of Unbalanced Production Lines Using Simulation Modeling: A Case Study for Solar Silicon Manufacturing [J]. Sustainability 4:697
- Su X, Lu J, Chen C, Yu J, Ji W (2022) Dynamic Bottleneck Identification of Manufacturing Resources in Complex Manufacturing System [J]. Appl Sci 12(9):419
- Kumar S, Petchrompo S, Ahmed T, Jain AK, Simulation-Based POLCA Integrated QRM Approach for Smart Manufacturing, Industry 4.0 and Advanced Manufacturing: Proceedings of I-4AM 2022. Lecture Notes in Mechanical Engineering, P421-434
- Dennis B, Daniel U, ,Andreas S et al Complex Job Shop Simulation CoJoSim—A Reference Model for Simulating Semiconductor. Manufacturing[J] Appl Sci 2023, 13(6):3615–3615
- P. Z L.Management Decisions in Multi-Variety Small-Batch Product Manufacturing Process[J]. Int J Simul Model 2022, 21(3):537–547
- Li M-H, Indriastuti S (2023) A system dynamics approach of a laptop manufacturing supply chain operation simulation modelling using AnyLogic, IET Conference Proceedings, Volume Issue 35 P210-217
- Uhlemann HT, Lehmann C, ,Steinhilper R The Digital Twin: Realizing the Cyber-Physical Production System for Industry 4.0[J]. Procedia CIRP,2017,61:335–340
- Tao Fei LW, Ran Z, Meng et al (2019) Digital twin five-dimensional model and ten domain applications_ Tao Fei [J]. Comput Integr Manuf Syst 25(1):1–18