So, in my facility we work with rolled plastic. For instance, we have finishing machines that take a roll of printed & laminated material as input and boxes of plastic bags as output. This machine will have a dozen bars spaced on it that heat, cut, compress, or otherwise interact with the material. Each time someone sets up the machine, they need to figure out where to place each of the bars and what temperature each of them should be… these things are different between one bag and the next and require time to get right. Time that the machine has to be running, churning through material we’ll never get to ship to the customer. What if we could tell the operator all these settings so they could get them right the first time?
We’re calling all these settings the run conditions. The run conditions that are temperature based will be consistent for bags that are constructed out of the same materials, even if the bag sizes are different. However, the run conditions that are positional based will be the same for any two bags that are the same dimensions, even if the bags are composed out of different materials. When I run a new bag A that’s the same material as old bag B and the same size as old bag C, the system should be able to tell it to use the temperature we used on B and the positioning we used on C.
Different types of bag zipper also result in different heat levels, even for the same material. And as we move to more machines we’ll see more constraints as well. On the laminators and slitters, the speed of the machine impacts the tension we want the plastic under. Information on the artwork affects the laser placement on the slitter, as well as the press machines. When all this is done, I expect we’ll have dozens of different possible inputs that can be used to group items together that share the same run condition value. The goal is to never have to give the system a value if the system can come up with it itself.
As I was on my way to work this morning, it occurred to me that this is somewhat like linear regression in machine learning. Linear regressions has a list of variables and a weight to each of them. The run condition definition is just like the list of weights, all of them 0 or 1. The run condition values are like the variable values. If we were able to run our data through a linear regression model, we could come up with the values for materials and bag dimensions that have not yet been run.
A pipe dream… for now. AX 2012 is 10 years old and has no machine learning capabilities. We’re replacing it in a year with Dynamics 365 though, and I suspect that it does have machine learning, for the master planning at the very least.
Joseph's Code of many colors 