[ot][spam] ai-guided 3d printing
how i found papers: i looked straight for papers, but no info. looked for news, which mentioned commercial products. i looked for the commercial products, but no info. i looked for papers that mentioned commercial products and found things this way. as if were 14 (i’m roughly 40) i showed off a gear assembly i 3d printed to my father. he said i should make a rocket like he had seen in the new. well, i’m all about shooting for the moon. the rocket from the news was ai-designed. due to dissociation, i’m only stealing mainstream approaches and not pushing the edge of algorithm training (until it’s more common to?) i’m presently experiencing inhibition (due to resisting it aggressively again) and have stopped playing video games [again?] so this seemed like something that reasonably crossed productivity and recreation. here’s a recent paper on ai and 3d printing: https://link.springer.com/chapter/10.1007/978-3-031-20875-1_33 Computer Vision Based Analysis for Fused Filament Fabrication Using a G-Code Visualization Comparison Abstract [edited down for conciseness and clarity] a probability of 41.1% remains that the printed part will have errors. This investigation provides an account of a camera based monitoring system developed to detect complex problems. - Image segmentation was used to remove the background of the printed part and the result was compared to a visualization of the G-Code. By using an exclusive-or method it was possible to determine differences, which can indicate defects. Depending on the similarity, the printing process can be canceled promptly. Tests have demonstrated that this method works reliably even under changing lighting conditions in most cases but can lead to poor segmentation due to shadows being cast in the infill. The application is also able to recognize differences when printed parts detach or layers have shifted if they are not covered by lower layers. The use of a light source on top of the 3D printer and additional cameras, beside the build plate, could solve both problems in the future. comments: addresses detection of failure, which is possibly a part of a potential larger problem of ai-designed g-code. the approach of image segmentation appears simple, one could probably finetune a popular pretrained model or 3d architecture to outperform this method. here’s another similar-looking recent paper: https://run.unl.pt/handle/10362/149946 Building a Fused Deposition Modelling (FDM) 3D Printing Visual Defect Detection System, Part 1: Creation of a Dynamic Imaging System, a Pixel-wise Segmentation Dataset with a Hybrid Synthetic Data Creation Method, and a Semantic Segmentation Algorithm with the SegFormer Deep Learning Model Abstract: As a part of an effort to develop a surface defect detection system for FDM 3D printed objects, this work project studies the application of the SegFormer network to semantically segment 3D printed objects. The project also showcases an affordable and accessible imaging system designed for the surface defect detection system, to support the decisions made during the segmentation task and to be used to evaluate the segmentation models. To achieve this, the first-ever pixel-wise annotation dataset of 3D-printed object images was created. Model-O1, a SegFormer MiT-B0 model trained on this dataset with minimal data augmentation resulted in an Intersection-over-Union score of 87.04%. A synthetic data creation method that caters to the nature of 3D printed objects was also proposed, which expands upon existing synthetic data creation methods. The model trained on this dataset, Model-A2, achieved an IoU score of 89.31%, the best performance achieved among the models developed in this project. During the evaluation of the model based on the inference results, Model-A2 was also identified to be the most practical model for building a surface defect detection system. comments: this paper references a 3d printed imagery dataset which sounds quite helpful if public. it’s a little confusing to me that it discusses “synthetic data creation” but is only addressing defect detection and not g-code synthesis. this is because i searched for the term “spaghetti detective” which is a corporate-built defect detector. the seed article i started with was https://www.hubs.com/blog/ai-assisted-3d-printing/ . it may be productive to look for other phrases from the article.
thought: ideally i would find the actual rocket design research, or research it cited there are kind of two failure points with printing: failure of the print, and failure of the part a strong design system would integrate both, such that design of the part is i informed by strengths of the printing process. the concept crosses physics simulation, materials properties, computer vision, defect detection if we narrow in on defect detection, which i recently glanced at, could we imagine stretching this to a simulation? how might a computer learn to simulate a 3d printer? this is something i’ve avoided learning about. for each step, it could be a big slow trained algorithm, or it could be a manually designed system, or it could be a hybrid where an AI builds a manually designed system for example: maybe it would be quite helpful to consider a system that builds a simulation of a mechanism based on observation of it. this has overlap with existing motion NeRF research, but is simpler because it would output a simplified vector description rather than a visual simulation.
here’s a recent paper mentioning generative design and g-code: https://link.springer.com/chapter/10.1007/978-3-031-24457-5_38 Study of Improving Spur Gears with the Generative Design Method Abstract [edited]] the question of how to improve the geometry of these parts was always in the open. idea of reducing the material volume used in the manufacturing process by using a generative design method, meaning a free-flowing design using generative design software. CAD model is subjective to parameters that are targeted in the program (volume, mass, safety factor) that are a frame of reference to develop different iterations from which we chose 5 of them. A finite element analysis is made to compare the generative version with the non-modified CAD to see the differences in behavior. These results are used to choose the best iteration of the gear. The parts modified as well as the non-modified ones are manufactured using different additive manufacturing technologies. An analysis is made to see the differences in terms of mechanical characteristics by making tests for surface durability, tensile strength, and bending. The results of the analysis show a 50% decrease in material volume without compromising the structural functionality. comment: this is cool, seems like print optimization. it sounds like the computer might not be intelligent here, simply trying out changing parameters? maybe i am wrong? but they managed to optimize and design and made have data. experiencing further inhibition. want to look up what finite element analysis is.
it looks like finite element analysis is meant to mean toolboxes that use a raster approach to analyse physical systems
thinking this finite element analysis stuff sounds great to have in a loop with 3d print designs. there must be open source libraries.
tools listed at https://en.wikipedia.org/wiki/List_of_finite_element_software_packages there are a lot of free ones it looks to me which means deciding. i wonder if any have ongoing open source projects associated with3d rinting or anthing?
this might have been my big milestone. - finite element analysis (FEA) is used for simulation of material properties. even with 3d printing. - there are papers under terms “generative design” “defect detection”. haven’t tried “print optimization”. the first one linked above has some print image data. others likely have more things. be nice to find associations of fea with 3d printing.
thinking briefly on concept of using fea software to train a model that quickly predicts relevant properties to a goal involved in generative design. this would help a system trying to make designs converge. i suppose usually that would simply be feedback into the generative design system.
it looks like there’s a lot of information on training models to perform fea tasks, from mainstream manufacturing. i found a paper that crossed some topics from 6 days ago but it was paywalled. having a new inhibition. likely time to do other things.
I was thinking of this and how a more reasonable goal than launching into space might be simply to make some automatic printer calibration code. It could use computer vision, or the user could even just enter the number of the best test print. Software like that can seem hard to find. I don’t have a 3D printer where I am right now.
note: printing is usually done via delayed g-code, there is more potential if the printer head is directed live by the design system
there’s a 3d printing protocol sugar library at https://github.com/rtellez700/mecode that looks like it would work for gcode generation or for live control the people currently maintaining it from harvard have a frightening website, it shows an unfolded box they 3d print as “living matter” that quivers, starts folding itself, flips itself over and keeps quivering like an injured insect they’re 3d printing biological hearts that beat on their own
simple 3d printer ideas that surprisingly may stimulate inhibition in me for innovation: - you could possibly extrude in a pattern such that the filament doesn’t bind to the layer below it, making support that vacuums or blows away with an arbitrary extruder and filament. expecting this to be slower from waiting for the filament to harden before letting it come in contact with the layer below. could be existing paper or software on trying something like it somewhere, but maybe not since it addresses low-end devices
AI-Calibration-and-Slicing idea: -> strap a camera to the printer head using zipties or a shoelace - print many tiny items in varying ways => use video-to-nerf technology to make 3d models of the small printed objects - compare the models to input data, and produce a high quality printing model
searched for “finite element” on github: Sort: Most forks dealii/dealii The development repository for the deal.II finite element library. c-plus-plus finite-elements 1.1k stars C++ Updated 44 minutes ago mfem/mfem Lightweight, general, scalable C++ library for finite element methods hpc scientific-computing high-performance-computing parallel-computing amr fem finite-elements computational-science high-order math-physics radiuss 1.2k stars C++ BSD-3-Clause license Updated 14 hours ago 2 issues need help
searched for “finite element” on github, surprisingly few highly popular results, maybe most use acronyms Sort: Most forks dealii/dealii The development repository for the deal.II finite element library. c-plus-plus finite-elements 1.1k stars C++ Updated 44 minutes ago mfem/mfem Lightweight, general, scalable C++ library for finite element methods hpc scientific-computing high-performance-computing parallel-computing amr fem finite-elements computational-science high-order math-physics radiuss 1.2k stars C++ BSD-3-Clause license Updated 14 hours ago 2 issues need help
participants (1)
-
Undescribed Horrific Abuse, One Victim & Survivor of Many