Can AI and 3D technology combine?
Nowadays, AI is leading the field of science. Machines can learn and complete tasks independently. However, 3D printing is much more than the production of plastic prototypes. 3D printing is still quite a complex process, and AI can improve it and make it even more efficient. There is a great chance that the combination of Artificial Intelligence and 3D printing will lead to new applications in the additive manufacturing technology field. This article is about the applications where AI can be used with growing 3D printing technology.
There is a question coming to my mind that What is the requirement that we need to combine AI and 3D printing technology? I mean AI is used everywhere from recommending YouTube videos to diagnosing the person; even more than that which results in every human being can carry out everyday tasks very easily. As I said earlier, 3D printing is quite a complex process. combining 3D printing and AI could increase the performance of a 3D printer by reducing the risk of error and facilitating automated production. Machine Learning is currently being used to solve the 3D printing problems by using generative design and testing in the pre-fabrication stage, with aim of improving cost savings and printing efficiency. In order to create intelligent service-oriented development processes for the industry, AI is currently finding applications in 3D printing and additive manufacturing (AM). How it can be combined? What are those tasks in 3D printing which require carrying out in an optimum way to reduce human errors?
Let’s dive into it…
Producing the defective 3d print output leads to a major loss in time, money, and more wastage of materials. So, We can have the Machine Learning algorithm which can continuously monitor 3d print and check whether the print generated is defective or not. If it is defective then stop printing else it’s good to go for the final output. You can clearly visualize that how hectic is that !!! The following image shows certain 3d printing defects.
I was going through some repositories. To do some hands-on, I have attached the GitHub repository link:
- https://github.com/alexgrigoras/failure_detection_for_3d_printing
- https://github.com/Mattalabs/OctoPrint-Mattacloud
- https://github.com/XiaoJiNu/DEye
- https://github.com/dnstanciu/pool-tracker-pro
- https://github.com/akornyukhin/3d_printing_defect_detection
You can clone any above GitHub projects or look my code 😁😎 too. All the steps to perform experiments are written in the ReadMe.md file in the every GitHub repository link. Although, experiments give more idea instead of reading stuff. 😎🤖
Generation of 3D Shapes from Free-Hand Sketches
What if we have only the handy sketch of the model that we want to print in 3D ? There is AI present to help you out from this situation. If we could convert any handy sketch to a 3D model, it would cultivate new way to create 3D Data for 3D Printing. There is special algorithm of machine learning called as GANs i.e., Generative Adversarial Network which can do the stuff that is convert the 2D sketch into 3D sketch. The basic pipeline is shown below that how the GANs generate the 3D shape is generated.
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).
Training GANs is very tedious, hectic and time consuming task.😤 Better to go for pretrained model.🤠
To do some hands-on, I have attached the GitHub repository link
- https://github.com/jpjuvo/64-3D-RaSGAN#tutorial
- https://github.com/msraig/mp_gan
- https://github.com/XingangPan/GAN2Shape
- https://github.com/zxpzhong/3D-RecGAN-pytorch
Hands-on Experiment
I have developed the demo of the surface failure detection on the steel materials. It has total 6 classes of defect called as Crazing, Inclusion, Patches, Pitted, Rolled and Scratches. I have trained the simple neural network on the dataset of steel images and I got 97% validation accuracy and the loss is below 0.5. You can also view my **code.**
Here are testing results on the neural network I have trained. In the below image, my neural network made only one false prediction i.e., image with label red is predicted false.
To clarify more in scientific way, on the basis of feature extraction of geometrical anomalies occurring in infill patterns due to inconsistent extrusion, weak infills, lack of supports, or sagging and compare it to the features of a perfect 3D print. This approach is built on the concepts of image classification and computer vision using machine learning, which is an extremely popular technology because of the availability of datasets, monitoring systems, and the ability to detect causal relationships of defects. To check the quality of the parts, an integrated camera with the 3D printer captures images at regular intervals and process it using the CNN model. The results of this project are a more optimized and automated 3D printing process with the potential to solve the most widespread problem of product variability in 3D printing.
Daghan Cam, CEO of Ai Build explained that “We have a simple rule at Ai Build: if a product is faulty, we repeat production. If it has a small defect, we repeat. If we are slightly in doubt, we repeat,” explained. “This level of perfection in additive manufacturing usually comes at a high cost, in the form of excessive labor and material waste.”