by Christophe Ponsard, Renaud De Landtsheer and Birgit Palm (CETIC)
Reflecting the state of a complex physical asset or process into its digital twin cannot be a perfect process. However, accurate reasoning must stay possible on a digital twin even in case of partial or temporary degradation of its connection with its physical counterpart. In the scope of an Industry 4.0 project, we are investigating how to deal with such a challenge for the optimised operation of a steel factory.
The digital twin concept can be defined, fundamentally, as “an evolving digital profile of the historical and current behaviour of a physical object or process that helps optimise business performance” [1]. The concept emerged from the growing digitalisation in many sectors like Industry 4.0 together with the big data technologies enabling the exploitation of the massive amount of data being generated. However, there are risks associated with the blind use of technology, such as the inability to know how close to reality the digital twin is at a given time, making it difficult to know what can be predicted by using it as a dynamically evolving model of reality.
In order to tackle this problem in the case of partially defined workflow, we developed an approach mixing both a flexible workflow model including physical constraints and a data-driven process mining based on various types of field sensors. The aim of such a twin is to be able to detect how well/optimally the process is behaving and in case of deviation to explore a way to restore normal operation with minimal impact.
Our context is the TrackOpt project aiming at improving the tracking and optimisation of steel making processes [L1]. The project involves both specialists in steel (the German BFI steel research centre and the Ferriere Nord company in Italy) and in optimisation (CETIC in Belgium and Scuola superiore Sant'Anna in Italy). Figure 1 shows an exemplary steel making process organised as a pipe of specific processing steps. A number of ladles are moving from one processing station to the next either using a crane or a ladle car (on rails). A number of waiting/swapping positions are also possible between stations. Of course many ladles are engaged simultaneously in order to ensure continuous casting.
Figure 1: Typical steel making process using moving ladles.
Building a digital twin combines a global workflow model with known constraints such as typical durations due to physical constraints (e.g. for melting, casting) and a data collection process to gather key information about the ladle position using weight sensors and manual encoding of identification. Increasingly automated recognition of ladle is also used and is one of the aims in the project TrackOpt. However the information might be partial or wrong, e.g. in case of encoding error, recognition failure (due to harsh environment/sensor malfunction). In this case, uncertainty is introduced into the digital twin which means the real process could be in a collection of alternative states resulting somehow in a multiplicity of digital clone variants. Accurate reasoning can, however, be maintained by applying the workflow model to each possible twin variant used as model, in a way close to the experimental twins approach [2]. In addition, we also apply a reconciliation process based on the available information which allows us to reduce the variant scope to the right alternative until the digital twin is unique and precise again. The exact meaning of accuracy can be defined based on a goal-based analysis of the required properties together with an audit of the system monitorability also coping with the occurrence of risks impacting it [3]. This processing could also cope with the occurrence of process deviations, like when a ladle needs a repair between stations and needs to be optimally reinserted to preserve the steel quality. In this case, the system can explore possible schedules and propose a corrective one even based on partial knowledge of the whole system. We plan to implement such an engine based on the OscaR optimisation library and more specifically the constraint-based local search engine [L2].
References:
[1] A. Parrott, L. Warshaw: “Industry 4.0 and the digital twin manufacturing meets its match”, Deloitte, 2017.
[2] M. Schluse, L. Atorf, J. Rossmann: “Experimentable digital twins for model-based systems engineering and simulation-based development”, IEEE SysCon, Montreal, QC, 2017.
[3] A. van Lamsweerde: “Requirements Engineering: From System Goals to UML Models to Software Specifications”,.Wiley, 2009.
Links:
[L1] https://www.cetic.be/TrackOpt
[L2] https://bitbucket.org/oscarlib/oscar
Please contact:
Christophe Ponsard
+32 472 56 90 99
