Every number below comes from the coaster itself: measured on the bench, or from the simulator that runs the coaster's real control code against a physical model fitted to those measurements. Swipe, or pick a topic.
Against a traditional on/off thermostat
对比传统开关式温控
The usual way to drive a cooling element is a thermostat: full power until the drink is cold enough, off until it has warmed a degree, repeat. Frostflare instead works out the smallest current that holds the temperature and runs exactly that. Getting there takes the same time. Holding it does not.
常见做法是温控开关：全功率开到够冷，关掉等它回温一度，再开。霜焰则算出「刚好守住这个温度」的最小电流，一直按这个电流跑。降到温的时间一样，守温的代价完全不同。
Simulated, 300 ml held at 15 °C, room 25 °C: power from the adapter
仿真：300 ml 保持 15 °C，室温 25 °C，从电源取的功率
Frostflare霜焰7.6 W
On/off开关式35.4 W
Measured on the production board, empty plate, room 27 °C: power holding 8 °C
实测：量产板，空盘，室温 27 °C，保持 8 °C 的功率
Frostflare霜焰12.5 W
On/off开关式28.5 W
Measured: holding 50 °C
实测：保持 50 °C 的功率
Frostflare霜焰3.8 W
On/off开关式7.6 W
Measured: temperature swing while holding 8 °C
实测：保持 8 °C 时的温度波动（峰峰值）
Frostflare霜焰0.4 K
On/off开关式3.4 K
- Half the power to hold a temperature, measured: 12.5 W instead of 28.5 W at 8 °C, 3.8 W instead of 7.6 W at 50 °C. With 300 ml in the simulator, holding a cool 15 °C takes 7.6 W instead of 35.4 W, and 50 °C 14.1 W instead of 20.7 W; getting there takes about the same time (15.0 and 14.4 min from 25 to 15 °C).
- Steady, not see-sawing: measured, the plate stays within 0.4 degrees instead of swinging 3.4 (5 when heating).
- No clicking on and off every few minutes, and the fan is not flat out the whole time.
- 保温只要一半的电（实测）：8 °C 时 12.5 W 对 28.5 W，50 °C 时 3.8 W 对 7.6 W。仿真里放 300 ml，保持清凉 15 °C 是 7.6 W 对 35.4 W，保持 50 °C 是 14.1 W 对 20.7 W；降到温的时间差不多（从 25 °C 到 15 °C 分别是 15.0 和 14.4 分钟）。
- 稳，而不是来回摆：实测杯底波动 0.4 度，开关式是 3.4 度（制热时 5 度）。
- 不会每隔几分钟咔哒开关一次，风扇也不用一直满速。
Simulation: 300 ml of water, the coaster's own control code, plant parameters fitted to bench measurements of the production board (33 µH). Measured: the same board, empty plate, each run started from equilibrium. On/off = full power, ±1 K hysteresis, fan at 100 %.
仿真：300 ml 水，杯垫自己的控制代码，物理参数按量产板（33 µH）台架实测拟合。实测：同一块板，空盘，每段都从热平衡开始。开关式 = 全功率、±1 K 回差、风扇 100 %。
A filtered, smooth current instead of chopped 12 V
平滑的直流，而不是被斩开的 12 V
The cheap way to vary a cooling element's power is to switch the full 12 V on and off fast (PWM). But the heat an element wastes in itself grows with the square of the current: pulses of 5 A half the time waste twice as much as a steady 2.5 A, while moving the same heat. Frostflare filters its drive through inductors and capacitors, so the element sees a smooth current.
调制冷片功率最省事的做法，是把 12 V 快速地开、关（PWM）。可制冷片自身浪费的热跟电流的平方成正比：一半时间 5 A 的脉冲，比一直 2.5 A 多浪费一倍，搬走的热却一样。霜焰的驱动经过电感和电容滤波，制冷片上是平滑的电流。
To draw 5 W out of a drink at 8 °C (heatsink 35 °C), the element needs
从 8 °C 的杯子里抽走 5 W（散热端 35 °C），制冷片要耗
Smooth current平滑电流11 W
12 V PWM12 V PWM33 W
Measured: power to hold near room temperature, 10 µH vs 33 µH inductors
实测：接近室温保持时的整机功率，10 µH 对 33 µH 电感
33 µH (production)33 µH（量产）0.84 W
10 µH (prototype)10 µH（样机）2.34 W
- A third of the power for the same cooling at partial load, which is where a coaster holding a drink spends its day. With the same average current, 12 V PWM would not cool that drink at all: it would warm it by 1.5 W.
- Kinder to the element: no thermal shock every cycle. Element makers ask for under 10 % ripple; chopped 12 V is 100 %.
- The filter was sized on the bench: the production 33 µH inductors cut the ripple in each from 12.5 A to 1.9 A peak-to-peak and halved the holding power (and the board's warmth by 10 K).
- 部分负载下同样的制冷只要三分之一的电——而守着一杯饮料的杯垫，一天里大部分时间就在部分负载。按同样的平均电流，12 V PWM 根本冷不下那杯饮料，反而往里加热 1.5 W。
- 对制冷片更友好：没有每个周期一次的热冲击。制冷片厂家一般要求纹波小于 10 %，斩波 12 V 是 100 %。
- 滤波器是在台架上定的：量产的 33 µH 电感把每路纹波从 12.5 A 降到 1.9 A（峰峰值），保温功率减半，板子也凉了 10 度。
First chart: TEC model with the element's measured constants (Seebeck 49 mV/K, 2.10 Ω, 0.68 W/K). Second: bench, same room, same hold, the two prototype boards side by side.
第一张图：按实测的制冷片参数（塞贝克 49 mV/K、2.10 Ω、0.68 W/K）计算。第二张：台架实测，同一房间、同一工况，两块样机并排。
It knows the room it is standing in
它知道自己在什么样的房间里
How hard the fan has to work, and how cold the heatsink may get before the air on it condenses, both depend on the room. Frostflare measures the room's temperature and humidity and adapts to both, instead of assuming a lab at 25 °C.
风扇要多卖力、散热片最冷能到多少而不让空气在上面结露，都取决于房间。霜焰测着房间的温度和湿度并随之调整，而不是假设自己永远在 25 °C 的实验室里。
Measured: room-temperature error with the screen at full brightness
实测：屏幕全亮时的室温误差
Compensated背光补偿0.3 K
Fixed offset固定偏差5.7 K
- Quieter when it can be: the fan runs only as fast as this room requires, not at a speed chosen for the worst case.
- No water on the desk: at 25 °C and 60 % humidity air condenses below 16.7 °C. While heating, the heatsink is the cold side; with condensation guard on (Settings › Temperature), Frostflare keeps it at least 3 K above the dew point, by easing the current rather than speeding up the fan.
- Calibrates itself: the humidity sensor sits on a warm board and reads about 6 K high, up to 5.7 K more while the screen is lit, and more again when the heatsink beside it is hot. Whenever the coaster rests it learns its own offset from its plate and heatsink sensors, and it takes the screen's and the heatsink's warmth back out as they change, so every unit reads the real room.
- 能安静就安静：风扇只转到这个房间需要的速度，而不是按最坏情况定一个转速。
- 桌上没有水：25 °C、60 % 湿度的空气在 16.7 °C 以下就会结露。制热时散热片是冷的那一面；打开防凝露（设置 › 温控），霜焰让它保持在露点 3 度以上，靠的是收一点电流，而不是把风扇转快。
- 自己校准：温湿度传感器在一块发热的电路板上，读数高约 6 度，屏幕亮着时最多再高 5.7 度，旁边的散热片热起来还会更高。杯垫每次静置时，都会用杯底和散热片两个温度传感器学出自己的偏差，并随时扣掉屏幕和散热片带来的热，所以每一台读到的都是真实室温。
Dew point: Magnus formula. The screen's share: bench, both prototypes, bridge off, backlight stepped 0–100 % (fit 0.26 K rms, the same on both boards). Offset and heatsink share: bench, checked against a second coaster resting in the same room, through rest, −10 °C and 28.5 °C holds (fit 0.22 K rms).
露点：Magnus 公式。屏幕那一份：台架实测，两台样机，功率级关闭，背光 0–100 % 阶跃（拟合误差 0.26 K，两台一致）。偏差与散热片那一份：台架实测，以同屋静置的另一台杯垫为室温参考，经过静置、−10 °C 与 28.5 °C 保温（拟合误差 0.22 K）。
It reads the element while it drives it
一边驱动，一边读懂这片制冷片
A cooling element is also a thermometer: switch it off for a few milliseconds and the voltage across it tells the temperature difference between its faces. Frostflare does this continuously, identifies each element's resistance, and commands a current, never a raw voltage.
制冷片同时也是一支温度计：关断几毫秒，它两端的电压就说出了两面之间的温差。霜焰一直在这样测，辨识出每一片的电阻，并且下达的是电流指令，从不直接给电压。
Current surge when switching cold → hot with 40 K across the element
两面温差 40 K 时从制冷切到制热的电流冲击
Current command电流控制≤ 5.0 A
Raw 12 V直接给 12 V6.7 A
- More power is not always colder. Past an optimum the element heats itself faster than it pumps; measured on the bench, full power left the plate 0.7 K warmer than the current Frostflare picks.
- No surge on reversal: the element's own voltage (49 mV per kelvin) adds to the supply when the direction flips. A raw 12 V would push 6.7 A, above the element's 5.8 A rating; a current command cannot.
- Every element is different: resistance and Seebeck coefficient are learned on each coaster, and the simulator checks that a ±20 % spread still cools with no false alarms. The same readings catch a shorted or disconnected element.
- 功率大不等于更冷。过了最佳点，制冷片自己发的热比搬走的还快；台架实测，全功率比霜焰算出的电流反而热 0.7 度。
- 换向不冲击：方向切换时，制冷片自身的电压（每度 49 mV）会叠加在电源上。直接给 12 V 会冲到 6.7 A，超过制冷片 5.8 A 的额定值；电流控制不会。
- 每一片都不一样：电阻和塞贝克系数在每台杯垫上自己学出来；模拟器验证过参数偏差 ±20 % 仍能正常制冷、不误报。同样的读数也能发现制冷片短路或断线。
Bench, production board, empty plate, current swept at 40 % fan. Surge: 12 V plus 40 K × 49 mV/K across 2.1 Ω.
台架实测，量产板，空盘，风扇 40 % 下扫描电流。冲击电流：12 V 加 40 K × 49 mV/K，除以 2.1 Ω。
Fast, quiet or frugal: what each mode really costs
快、静、省：每种模式实际付出什么
300 ml from a 25 °C room to a cool 15 °C. Eco takes a little longer and uses about 24 % less energy getting there; Quiet caps the fan at 2500 rpm, the quietest, and is only a few minutes slower. Once there, every mode holds on about the same power.
300 ml 从 25 °C 的室温降到清凉 15 °C。节能稍慢，降温用电少约 24 %；静音把风扇限在 2500 转，最安静，只慢几分钟。到温之后，各模式保温的功率差不多。
Time to 15 °C
降到 15 °C 用时
TurboTurbo14.9 min
Standard标准16.6 min
Quiet静音18.7 min
Eco节能20.4 min
Energy to get there
降温耗电
Eco节能7.2 Wh
Standard标准9.5 Wh
TurboTurbo9.5 Wh
Quiet静音9.9 Wh
- The screen states each trade-off before you pick it, and those sentences are worked out from the same model as these charts.
- Turbo is only a little faster here: the fan is near its limit already at Standard's cap, so the current, not the fan, does the work. Turbo's extra is for a hot room or a big cup.
- 选之前屏幕上就写着每种模式的取舍，那几句话和这些图出自同一个模型。
- 这里 Turbo 只比标准快一点：风扇在标准的上限时已接近极限，出力的是电流而不是风扇；Turbo 多出来的那一点，是给热天和大杯准备的。
Simulation (固件计划 §7.1), 300 ml, room 25 °C, target 15 °C; Standard at 3300 rpm.
仿真（固件计划 §7.1），300 ml，室温 25 °C，目标 15 °C；标准按 3300 转。
IVY Drive
