Adam

Saw the buzz about a radiator design that we tested for an upcoming orbital data center demo mission. It’s not a conventional design, and also not one that’s intuitive, so why did we pick it? Several reasons: 1. First and foremost, we had to work within the confines of an existing ESPA-class design, and already had one hand tied behind our back when it came to expanding surface area. We’re a small startup, and can’t afford to upgrade to a bigger craft. With this in mind: 2. Theoretically, the radiation heat transfer equation is: Q=VF*σ*ε*A*(Tsurface^4 - Tspace^4), which relies on the delta between Tsurface and Tspace. The larger that delta is, the larger Q (heat transferred) is. If those numbers are the same, or close to the same, then Q is small. With multiple GPUs spread out inside and sensors in the center, then theoretically each fin would not be that much different in temperature from one another, thus minimizing the delta. 3. Yes, the view factor is poor (VF, a unitless range between 0 and 1 with 1 being perfect VF) to space, but they still have some VF to space. This means the longer and closer the fins are placed together, the more inefficient they become, resulting in diminishing returns (due to non-ideal VF). It’s also why a totally flat, geometric radiator with no fins is ideal. But we didn’t have that option due to 1. 4. The side fins are the least efficient, but still useful as a mechanical mounting point. We could honestly delete a few as we didn’t really need them.   So why didn’t we think the radiator design would cause thermal issues? It would require a significant delta-T between the fins. This would be a legitimate issue if we had all our heat sources biased towards a face, but this isn’t the case. But, theoretically, with our evenly distributed payload, if one fin is at 50C, then it's likely that the two fins on either side of it are also at the same temperature. So if the delta-T was at zero in the equation Q=VF*σ*ε*A*(T_surface^4 - T_surface^4), then Q should also be zero. Lastly, this also works in Ansys Thermal Desktop. Maybe we found a glitch in Thermal Desktop, I don’t know. But faced with the first constraint, the lack of alternative options, and the requirements of a GPU-dense payload, it was worth the prototype to find out.

make reality your bitch

10-32 has the mandate of heaven

Solidworks has surpassed Siri as the worst software I have ever encountered. It's literally just not capable of simple math. Insane.

I recently learned about the existence of calibrated rulers