To build a process model one does require a subject matter expert that understands the manufacturing process. Based on his input a set of values – temperatures, etc are identified.
Experience shows that, not everything that was initially identified is relevant and typically not all of the relevant values were identified initially. This means that LDTs for a specific manufacturing process are developed in stages.
Three major stages can be identified:
- M0 Initial assumption – based on the subject matter input – we learn what values are relevant.
- M1 Data modelling – cohesive data model created
- M2 Statistically verified – using analytical tools enables us to identify which values are relevant
- M3 Refinement – improvements based on actual usage of an M2 model
In this modelling process one of the critical questions is to what extent will this model be supported by already existing implementations (applications). For instance an ERP system was implemented with a certain process model assumption. There is a set up that maps the material flow, the operators output and the quality aspects of the production.
A lean digital twin does require specific information about the material flow and the operators output. Based on the type of the product this may requite product specific information.
Once the data modelling stage (M1) is completed, the model has to be compared with the ERP system in order to determine what data is already available. Data items not available in the ERP system will have to be collected independently. There is a specific part of the IoT scan that deals with these requirements.