The Open Factory Twin (OFacT) project aims to provide a digital twin for production and logistics environments. Digital Twins (DT) represent their environment as a virtual model of all relevant parts of the real system. OFact is meant to support the design, planning and operation control of discrete material flow systems and thus supporting management of the system during the whole life cycle.
Coming from the challenges such as ...
- shorter production life cycles
- frequently changing demands
- complex supply chains
- increasing number of possible product variants
- regulations, legal requirements and restrictions
... companies are faced constantly with complex decision that often needs a dynamic evaluation and comparison of various
scenarios. Often detailed simulation models are the only way to get a reliable evaluation of costs and performance. Data
of the real world has to be integrated regularly into the simulation models to keep them up-to-date.
In the design (or re-design) phase of a production system, the digital twin can be used to simulate different design
alternatives and evaluate them even before the real system exists. When the real system is in operation, the digital twin
can be used in an iterative way between planning orders and resources and controlling the plan during operations dealing
with disruptions in real time.
OFacT is based on a general state model that describes the state of the factory and the possible behaviors and can be used for all kinds of discrete material flow systems (e.g. assembly lines, flexible matrix production, job shops, warehouses or even supply networks). The model consists of the following basic elements:
- orders (that describe the "customer" demand)
- entities (resources and parts) that describe the physical objects in the system with parts being transformed based on processes that are executed by reusable resources
- processes define the possible transformation of parts (and sometimes resources) in time, space physical attributes and quality
While the processes describe the possibility space of the material flow system, (planned and actual) process executions describe the concrete transformation in the past, present and future and can be seen as event logs that capture the dynamic behavior of the system. Process executions can be created by the real system - planned process execution are generated based on data from planning systems such as ERP or APS systems while actual process executions are generated based of sensor or event data. Planned process executions can also be created by the multi-agent system that controls the state model. The control logic of the digital twin is realised by agent behaviours and thus separated from the static state model allowing for complex and flexible control behaviours. The separation of possibility space and actual behavior as well as the separation of state mode and agent-based control allows for a flexible and modular design of the digital twin that can be adapted to the specific requirements of the production system. Even more it facilitates the learning of the model from the data of the real system.
The OFacT framework consists of the following super components:
- Environment: provides components to interact with all kinds of environments (real and virtual) and to integrate data
- Data Integration: provides tools to integrate data into the digital twin including consistency checks and update mechanisms
- Work Instruction: provides tools to pass work instructions (planned process executions) back to the physical world (closed loop system)
- Simulation: is a virtual environment, that mimics the behavior of the real world and can be used to evaluate different scenarios or produce forecasts
- Data Space Connector: provides the tools to connect the digital twin to the data space and to share the digital twin with other companies
- Digital Twin: provides the state model, the agent control
- State Model: describes the state of the factory and the possible behaviors as well a passed and planned transformations
- Agent Control: provides the control logic of the digital twin based on order and resource agents
- Planning Services: provides tools to generate the state model and to create and evaluate scenarios
- Scenario Generation: provides the capabilities to create difference scenarios based on manual parameter variation, optimization or even Artificial Intelligence such as reinforcement learning agents or generative models
- Scenario Analytics: provides the tools to determine KPI's based on the state model, visualize and compare them
This project uses a monolithic repository approach and
consists of different parts that are located in different subfolders of the ofact folder.
Examples are use case-specific models and adaptions (currently only the twin models)
are offered in the projects folder.
Detailed getting started guides are described for every component in their dedicated README
file, located in the corresponding subfolders.
The following things are needed to run this application:
- tested on Python 3.10 and 3.12
- requirements.txt
The first release of the open factory twin contains the data model (state model component)
and therefore next to the agent control, the core element of the digital twin.
The state model can be filled with two sample use cases that can be found in the projects folder:
The models are provided in the {project_name}/model/twin/ folder, modeled in Excel files.
The template shows a small example shop floor that contains the most elements of the state model. In this example, a box is transported through a conveyor belt from the warehouse to the assembly station. At the assembly station, a worker processes the box content and the box content is transported back to the warehouse by an automated guided vehicle.
A more advanced scenario in the context of Industrie 4.0 is offered with the bicycle world.
Here, a modular and flexible assembly produces customized bicycles.
The assembly stations can execute one or more processes (standardized),
and the main product has a flexible assembly step sequences (routing flexibility),
restricted only by the assembly priority chart of each product.
This projects contains two variants, one with and one without material supply.
As stated before, the first release, contains only the data model (state model). However, soon further parts of the project will become open source. The aim is to offer an example case (bicycle world) that can be simulated (agent control) and analyzed (scenario analytics).
Contributions to this project are greatly appreciated!
For more details, see the CONTRIBUTING.md file.
This work is licensed under the Apache 2.0 license.
See LICENSE file for more information.
In the meantime, the project was created within the scope of the Center of Excellence Logistics and It - Nationales Leistungszentrum Logistik und IT.
-
Christian Schwede (HSBI | Fraunhofer ISST)
-
Jan Cirullies (FH Dortmund | Fraunhofer ISST)
-
Adrian Freiter (Fraunhofer ISST)
-
Roman Sliwinski (HSBI)
-
Jannik Hartog (Fraunhofer ISST)
-
Niklas Müller (Fraunhofer ISST)
If you have any further questions, please do not hesitate to contact us:
The documentation part of this work is licensed under the CC-BY-4.0 while the software part is licensed under Apache 2.0.