Simulation modelling/ Digital twin of Injection molding process?

Question

How can we make a digital twin of the Injection molding process in a production line? My idea is to do predictive maintenance and predict following downtime based on historical time series data of the Injection molding machine. Is Any Logic software capable of this operation?

Answer 1

Strictly speaking a digital twin would be a 1:1 copy of real-life production into some kind of simulator, which, by definition, is real-time (or can at least be slowed down to real time and NOW).

The beauty of the twin-idea quickly vanishes, when understanding:

• the sheer amount of required parameters needed for mapping (there is the effect of being too simple, see Ashby)
• their accuracy, availability etc. (e.g. how do you know wear-off from mold-to-mold or injection-to-injection sequence?)
• limitations of models, simulators, computers, people etc.

Also you'd need to know, for how long such predictions should be valid and reliable, e.g.:

• just for this mold, which will be changed in Q3 this year
• for any mold your company will ever do for the next decade
• and so on

It may turn out, that for predictive maintance you need and can use a very different set of parameters than having a 1:1 copy of your production line. The twin may even not provide such parameters.

AI-approaches are contemporary, and, when you look closer into it, are what the name suggests:

• Artificial Incompetence (by results)
• Acquired Incompetence (by this very technology)
• i.e. those systems "don't get a clue, don't get enlightment"

(Please get me right, I'm not against AI, but would like to open eyes about these "miracles".)

Turns out similar problems were already solved before products like you mentioned became avaliable. But this would be a bit out-of-scope for your question.

Answer 2

General numerical simulation packages should be able to do injection molding simulation (there may also be some modules especially for digital twins). For example, ANSYS has Polyflow module for this. As for the parameters, I would start with parameters that you already change or measure in the actual process.

Edit: This answer was focused solely on the digital twin. However, if your primary goal is to do predictive maintenance based on operation history, you don't require a digital twin. Instead, you should focus on data analysis, identifying key factors leading to downtimes and tuning their contributions over time. Only if such approach is not sufficient, digital twin might help you to further improve the predictive model.