Choose by power class and run length, not by preference. Below the 300 W class, with runs of a few hours, forced air is sufficient and simpler to live with. At higher power and for runs measured in tens of hours, a liquid loop keeps the lamp and optics steadier. A 300 W xenon source delivers 50 W of total optical power; the balance leaves the housing as heat.
Lamp power sets the size of the heat load; run length sets whether the housing ever reaches a steady state. Noise, layout and service access only refine that decision.
Power matters because optical output is a small fraction of electrical input. A lamp adjustable over 150–300 W delivers 50 W of total optical power, and a 400 W source delivers 100 W, of which 55 W is visible and 6 W ultraviolet. The remainder becomes heat inside the housing, and the cooling path decides whether it leaves through the fins or the arc chamber.
Run length matters because a housing is a thermal system with a time constant. A run that ends before the housing stabilizes never sees the drift; a run lasting all night sits at equilibrium for hours, and any weakness appears as a slow decline.
The difference is not the medium. It is where the heat goes, what noise and vibration the path adds, and how it fails.
| Criterion | Forced air | Liquid loop | Effect on your data |
|---|---|---|---|
| Heat path | Fins and forced convection to room air, exhaust ducted out | Cold plate to a chiller or facility loop | Air couples room temperature into the housing |
| Practical band | Lower lamp powers, intermittent runs | Higher power, runs of many hours | Larger input means more air to move quietly |
| Noise | Broadband fan noise at the bench | Pump and compressor, sited outside the room | Noise is also a vibration source |
| Vibration | Fan imbalance through the housing | Pump pulses through hoses; a remote chiller decouples | Beam position drift over hours |
| Installation | Duct to outside air or an extraction hood | Coolant, hoses, drain, footprint | Air is easier to install, harder to keep stable |
| Failure signature | Gradual drift as the filter loads | Step or alarm when flow falls | Two signatures, two responses |
Read the table by column. The air column assumes a room cooler than the housing and a duct that extracts; the liquid column assumes flow, coolant chemistry and a chiller that holds setpoint. Neither is better — you are choosing which failure you can detect earlier.
Forced air is right when the lamp sits in the lower part of the power range, the room is air-conditioned, and the exhaust can leave the room.
Three conditions have to hold. The intake must see clean, cool air; a source beside an oven or an open solvent bath pulls in warm, loaded air and loses its margin in summer. The duct must be sealed end to end, because air takes the shortest path and a gap at a collar sends most of the flow around the lamp rather than across it. The filter must be treated as a consumable, because a matted filter cuts flow and the housing runs hotter with no alarm.
Nothing can leak: no coolant, no hoses, no drain, no condensation. In a shared instrument room, that simplicity outweighs the extra thermal margin a loop would give.
A liquid loop is right when the source must hold output for many hours, when the housing is enclosed or racked with other heat sources, or when the room cannot be kept cool. Water carries heat out of the room, which matters when several instruments share a small space.
The loop also buys thermal isolation. A cold plate holds the lamp base at a set temperature rather than at whatever the room happens to be, and that removes one drift path. The cost is a new set of failure modes: flow restriction, depleted inhibitors, air pockets after a refill, and condensation.
If the coolant setpoint sits below the room dew point, water forms on the cold plate, the window and any unmasked metal. Water-cooled LED heads are specified over a −10 to 80 °C coolant range, and a setpoint near the bottom of it will condense in a humid lab. Stay above the dew point or purge the housing with dry air.
Both paths put a moving machine next to an optical bench, so site both as if the bench were the sensitive part. Fan imbalance produces a low-frequency hum that reaches the table through the floor; a pump produces a pulse that reaches it through the hoses. Both can disturb anything that integrates a signal over hours, though a stirred slurry reactor is rarely affected.
The measures are cheap: chiller on the floor rather than the bench, a service loop in the hoses, supported duct runs, and a beam check against a target at the start and end of a long run. If the beam has moved, the cooling path is one candidate alongside thermal expansion of the bench.
Either way, log room temperature beside the instrument log. When irradiance drifts, the first question is whether the room drifted with it.
Not practically. The housing, the cold plate and the airflow path are designed as one system, and a retrofit adds a plate to a housing that was never meant to conduct heat into it. The result is usually worse than either original design. Treat the cooling method as a purchase-time decision.
It removes one drift path, not all of them. A loop holds the lamp base and the optics at a steady temperature, so the decline caused by a housing heating up disappears. It does nothing for a lamp near end of life or a supply that cannot hold its current, and those leave a different signature.
Log irradiance for a typical run and compare it with room temperature. Marginal cooling produces a decline that follows the room and recovers after a pause at the same power setting. If it does not recover, look at the lamp and the supply before the airflow.
