Infographics of the intellectual system “i-feature” for processing typical elements in details
Yesterday I made a small infographic of my development (read: habrahabr.ru/blogs/artificial_intelligence/135795 )
Sorry for non-professionalism - just learning to design in that direction. It has long been wanted to start somewhere.
I present the perspective of the development of my future system in the context of using Cloud technologies and data storage and processing systems.
')
Briefly explain under the cut.
Divided into three parts:
1: collecting data from users and transferring them to servers for processing. There are formed training and test samples. Also, the sample is updated in the same place for retraining (or parallelism of two neural networks is used for more optimal and operational retraining)
- it is worth noting separately the development in perspective of tools for analyzing a solid model, checking for thin walls, curvature of the surface, etc. For some reason I am sure that this is real, because each surface in the model has its own mathematical description in space. This data also goes to the clouds to train the corresponding neural networks.
2: Actually developed technique. In this case, “i-feature” works on the principle of client-server. Those. all the math is rendered to the clouds, the subroutine accepts answers and sends requests. The linear algorithm works in a similar way.
3: The stage of final testing by the user of the results of the algorithm. Everything is reduced to the minimum parameters, which greatly simplifies the work with the system (especially for beginners). A tool is either created parametrically (by the way, I thought up how to integrate a neural network there), or takes it from an already created tool library. Also worth noting is the very real possibility of creating a single tool library for different manufacturing companies! Such globalization. Difficulties there are no special - mostly mechanical work. The parameters of the tools in the library can be used to train individual networks, which will greatly assist in the parametric creation of the tool!
Next, clicking a couple of buttons "OK", everything goes into the mode of regular work of CAD (generation of trajectories for the approved parameters and post-processing)
As promised, I am working, constantly improving the idea and algorithms.
Thanks for the feedback in the previous post!
Sorry for non-professionalism - just learning to design in that direction. It has long been wanted to start somewhere.
I present the perspective of the development of my future system in the context of using Cloud technologies and data storage and processing systems.
')
Briefly explain under the cut.
Divided into three parts:
1: collecting data from users and transferring them to servers for processing. There are formed training and test samples. Also, the sample is updated in the same place for retraining (or parallelism of two neural networks is used for more optimal and operational retraining)
- it is worth noting separately the development in perspective of tools for analyzing a solid model, checking for thin walls, curvature of the surface, etc. For some reason I am sure that this is real, because each surface in the model has its own mathematical description in space. This data also goes to the clouds to train the corresponding neural networks.
2: Actually developed technique. In this case, “i-feature” works on the principle of client-server. Those. all the math is rendered to the clouds, the subroutine accepts answers and sends requests. The linear algorithm works in a similar way.
3: The stage of final testing by the user of the results of the algorithm. Everything is reduced to the minimum parameters, which greatly simplifies the work with the system (especially for beginners). A tool is either created parametrically (by the way, I thought up how to integrate a neural network there), or takes it from an already created tool library. Also worth noting is the very real possibility of creating a single tool library for different manufacturing companies! Such globalization. Difficulties there are no special - mostly mechanical work. The parameters of the tools in the library can be used to train individual networks, which will greatly assist in the parametric creation of the tool!
Next, clicking a couple of buttons "OK", everything goes into the mode of regular work of CAD (generation of trajectories for the approved parameters and post-processing)
As promised, I am working, constantly improving the idea and algorithms.
Thanks for the feedback in the previous post!
Source: https://habr.com/ru/post/137396/
All Articles