👋 Happy Saturday! This weekend’s edition includes a look at a new method for generating structured 3D models from a single photo, a field guide to estimating transmission voltages by counting insulator discs, and a breakdown of why uranium enrichment is the bottleneck in nuclear proliferation.
Interesting Paper: One Image In, Structured 3D Out
ByteDance, Carnegie Mellon, and Peking University just released PartCrafter, a diffusion model that turns a single photo into a fully part-segmented 3D model.
For training, they pulled 3D models from Objaverse, ShapeNet, and the ABO dataset (standard sources in computer graphics research, but not something you'd see in a design or manufacturing workflow). These are mostly artist-created meshes with no feature trees, constraints, or considerations for manufacturing but often come with metadata that defines part groupings. Prior work discarded that structure and treated each model as a single mesh. This model retained the grouping, which turned out to be critical for learning what counts as a real, separable part.
As for engineering workflows: image-to-CAD models still don’t capture design intent as the models don’t understand constraints or how parts relate functionally, like in building a feature tree or defining mating conditions. Still, this is one of the first signs that part-aware generation might move beyond surface geometry and toward something usable in real CAD workflows.
Interesting Chart: Identify Power Line Voltage by Disc Count

A way to identify voltage levels in transmission lines: count the insulator discs to estimate how much energy is moving and what kind of system you’re looking at. Each disc typically handles about 10–15 kV, so a 230 kV line might use 15–20 discs. It’s not precise, but often enough to place the line in the right tier.
Lower voltages (under 100 kV) tend to support local distribution; 115–230 kV lines are regional; 345 kV and above are used for long-haul, high-capacity transmission. If you're walking around the neighborhood, you’ll see short poles with post-style insulators on residential distribution lines, not the long strings of glass discs used for high-voltage transmission.

Interesting Video: Barriers to Nuclear Proliferation
The bottleneck in nuclear proliferation isn’t bomb design, but uranium enrichment. Natural uranium is mostly U‑238 which can’t sustain a chain reaction, while the fissile isotope U‑235 (can undergo fast fission and trigger a chain reaction) makes up less than 1% and differs only slightly in mass.
Separating them becomes a brute-force problem solved through scale: build infrastructure that exploits the 1.3% mass difference. Two primary methods exist:
Gaseous Diffusion Plants: Uranium hexafluoride gas is pushed through porous membranes, where lighter U‑235 molecules diffuse slightly faster. This is repeated across ~4,000 stages to reach ~90% enrichment. The original Manhattan Project production facility scaled this up with over 100 miles of piping across 40 acres.

Centrifuge Arrays: Modern enrichment uses high-speed centrifuges. As the gas spins, U‑238 concentrates near the rotor wall while U‑235 remains closer to the center. Each stage increases enrichment by less than 1%, so facilities run up to 60,000+ units in cascade.

Manufacturing & Startup News
More leftovers from our weekly research, with a focus on global markets:
Avicena raised $65M in Series B funding to produce microLED interconnects for AI datacenters.
Voyager Technologies raised $383M in an IPO to advance its space and defense tech.
Phoenix Semi Corp raised $5M to recreate legacy semiconductors for obsolete supply chains.
Proxima Fusion raised €130M in Series A to develop a stellarator-based fusion power plant by the 2030s.
NuuEnergy closed a seven-figure pre-seed to boost heat pump adoption in Germany.
Co-Power raised €6.4M in seed funding to cut industrial energy costs with battery and solar systems.
Lendurai secured €5.6M in seed funding for GPS-independent UAV solutions for defense and commercial use.
Voltfang raised €15M in Series B funding to expand second-life battery storage to 250 MWh by 2026.
Refer a Friend, Earn Rewards 🎁
Refer just one friend to Hardware FYI, and we’ll send you our complete database of resources—every design guide, article, white paper, and more.
Thanks for reading to the end - if you’ve enjoyed the mechanics of these insights, consider sharing this issue with a fellow enthusiast!
After reading, I know at some point today I will find myself looking at electricity pylons with renewed interest :)
Good content!