The Data Centric Architecture of a Factory Digital Twin
a constraint . An FDT designed to handle processing materials that are liquids , slurries , or powders must account for the availability of a storage vessel specifically designed and connected for that purpose . Material oriented FDT must also manage cases where material can only be stored for a limited time , must be stored at least for a specified time , cannot be mixed across batches , or cannot be stored at all . In such processes , managing storage is of paramount importance .
In addition to recipe , equipment , and renewable resource information , specialized manufacturing processes have process physics specific data . For example , in biologics cell culture processes there are specialized cleaning constraints necessary to prevent pathogen contamination . For real world FDT development , the RTN consists of the general information described above plus the specialized data that the FDT must accommodate , which can be highly industry specific . As another example , consider coffee roasting . Once the beans are roasted a certain amount of degassing time is required by the recipe to provide high quality product . A coffee manufacturing FDT must satisfy this degassing time . Honkomp et al 2000 [ 8 ] provides a description of a variety of process physics that must be addressed by FDTs .
An important step in developing an FDT process model is to construct the appropriate RTN . This defines the level of detail that the model will be able to express and the data that is needed . A model with insufficient detail will result in unrealistic results , e . g . the model may generate schedules that cannot be executed with the real-world equipment in the process . A typical example would be making a liquid at 8 AM and consuming it at 10 AM when there is no tank available to store it during those two hours .
Too great a level of detail can result in a model that is slow to solve and results that are overly complicated and difficult to understand . As with all modeling activities , much of the art involved in building an FDT model is related to the critical choice of what is to be included and what is to be omitted . In addition , the RTN can evolve over time to increase or decrease the amount of detail according to changes in business requirements . In fact , this flexibility to easily change is one of the advantages of an FDT based on an RTN .
Note that the RTN based FDT describes the manufacturing process at a higher level and does not usually calculate the detailed physics of individual activities / tasks . For example , if a task represents mixing input materials , the FDT does not calculate the liquid flow field in a tank or calculate the degree of mixing as a function of time . Rather , the RTN is specified in terms of how long the mixing usually takes to go from pure inputs to the desired mixed output , perhaps with some modeled process time variability .
For purposes of understanding a mixing step in detail a computational fluid mechanics based DT might be developed . For a chemical reactor , a kinetics model – how fast reactants are converted to products - may be the basis of the activity / task digital twin and supporting the kinetics model might be a quantum chemistry model to gain insight into how reactants are converted to products . In fact , these examples show that real world artifacts – a factory – often need to be represented by a hierarchy of DTs .
Journal of Innovation 103