Deep Learning for Advanced Additive Manufacturing
3D Printing
Additive manufacturing, or 3D printing as it is often called, is a manufacturing process that has been developing steadily since 1984. Founded by Charles Hull, the process allows three-dimensional objects to be printed from digital data.
The global 3D printing industry is expected to reach $5.2 billion in 2020 and the largest opportunities for corporations in near future will be in home improvement and the life sciences sectors. Yet, today there are still several hurdles in the additive manufacturing process that need to be overcome for widespread adoption in the industry.
For example, additive manufacturing involves numerous and complex variables to be monitored and controlled in the process to achieve an acceptable level of accuracy in printing. Trial and error methods for finding the correct lattice positions or design of appropriate support structures are not a sustainable or fast solution.
AI Applications in 3D Printing
Machine learning is currently being used to solve 3D printing problems by using generative design and testing in the pre-fabrication stage, with the aim of improving printing efficiency and cost savings. Artificial intelligence is currently finding applications in 3D printing and additive manufacturing for creating intelligent service-oriented production processes for the industry.
Example : Defect Detection
General Electric’s GE labs in upstate New York claims to have previously developed computer vision technology that can find microscopic cracks in machine parts and other microscopic aberrations. The Additive Research Lab at GE Global Research in Niskayuna, New York claims to be using artificial intelligence and machine learning, to enable 3D printer to perform inspection of parts after they are completely built in order to improve cost and time savings in the manufacturing industry.
The team from the additive research lab uses high-resolution cameras to film the printing process for each layer to record streaks, pits, divots and other patterns in the printing powder which are invisible to the naked eye. GE claims that a proprietary machine learning platform then matches recorded powder patterns to defects revealed by CT scanners. In essence, the ML platform is trained through the use of high-resolution camera footage and CT scan data and can eventually ‘learn’ to predict problems and detect defects in the printing process.
Generative Adversarial Networks
Generative Adversarial Networks (or GANs for short) are one of the most popular Machine Learning algorithms developed in recent times. They are deep neural net architectures comprised of two nets, pitting one against the other (thus the “adversarial”).
GANs’ potential is huge, because they can learn to mimic any distribution of data. That is, GANs can be taught to create worlds eerily similar to our own in any domain: images, music, speech, prose. They are robot artists in a sense, and their output is impressive — poignant even.
ShapeNet Dataset
ShapeNet is an ongoing effort to establish a richly-annotated, large-scale dataset of 3D shapes.
It covers 55 common object categories with about 51,300 unique 3D models in different formats.
We convert all the STL data into 3D Voxel Index (Voxel is a 3D pixel, thus 3D Voxel file is similar to 3D bitmap file). It is easier to apply Voxel Data to deep learning libraries, because deep learning libraries are basically implemented for image processing based on bitmap (photo) file format.
We can train GANs by using The ShapeNet dataset. As a result, our machine learning system could become to create new 3D Voxel file(as an output) from a new 2D profile image (as an input).
A handy 2D sketch of a 3D model is one of ambiguous information for the machine.If we could convert any handy sketch to a 3D model, it would cultivate new way to create 3D Data for 3D Printing.
