Feeder Simulation / Digital Twin

Parts in automation generally have more predictable behavior – once we gain control. The first principle of automation design is to never lose control of the part. Feeders are the entropy fighters that deliver that control.

Unlike assembly automation, feeding has historically relied on direct testing and iteration (plus often rules of thumb and a bit of black art), rather than formal engineering methods.

In the last decade it has become feasible to simulate this randomness of bulk part behavior with a sufficiently large number of rigid bodies. This is invaluable to assess feeding concepts prior to the tooling build, when customer parts may not even be available in test quantities. In addition, limit conditions can be challenging to imitate in real-world testing. For example: What happens if the friction of the parts changes over time? Will the parts jam if they are all at the lower end of the tolerance?

We have adapted this DEM simulation to emulate the sensing and logic of flex feeding. The code can determine which parts are pickable within the camera’s field of view, estimate the time of the robot when a pick is available, and remove the part from the simulation. This allows us to develop part conditioning tooling to optimize the spacing and orientation of the parts. Receiving nest geometry can be tested and optimized against a large virtual variability of presentation, while still datumming the parts for downstream stations.  Flex feeding is an inherently statistical process – now we can check the tails.