Since most of us are trapped in doors with loads of free time or in my case hours of being trapped in my own headspace I’ve decided to do something outside the box. I’m going to start a project that attempts to explore some areas of computer automation using machine learning to optimize not only cam operations for a given project, but also explore using machine learning techniques to tease out optimal speeds and feeds for better finishes while still maximizing MRR. I’ve thought about this for a while that it seemed odd that I have to still tell the cam package what I think the optimal operation combination is to mill out my parts, and why for the most part the software can’t look at my model, available toolpaths and a list of tools I have on hand + a combination of constraints based on shapeoko limitations and give me a setup that it thinks is best for what I want to mill. For now I can build the predictive models, run the simulations and use run time as the main goal to start, and then validate the best predictive model on the shapeoko to see how it fairs against something I cam myself. Anyone think this is a cool idea, or has cabin fever consumed me?
so I built basically a cam library and a tool to use that… so you don’t have to invent everything
soon I want to build a tool to evaluate forces during the cut, that could be a judge of “goodness” in addition to cut time…
see https://github.com/fenrus75/FenrusCNCtools for info
What do you mean by cam library? At some basic level I have all of the Cam operations available to me in Fusion360 + tools available in my current tool library. The restriction is 3 axis milling, travel limits of the shapeoko, speed/feed restrictions based on tool and makita router and current stiffness of the setup with consideration to material being cut. At this point its bit non deterministic sense Fusion determines the toolpaths for the ops based on a lot of factors, but some ops are better than others depending on the stock remaining and tools in inventory. I’m just mapping all the features for now and making decisions about how to simplify the approach and see how good of a predictive model I can build and how it scales against more complex milling operations.
This topic was automatically closed after 30 days. New replies are no longer allowed.