How Air Keeps Getting Into a System That Is Supposed to Be Closed

Why this matters

"No leak found" is the most common conclusion on an air ingress call and it is usually correct, which is exactly the problem. Half the routes air uses to get into a closed system do not lose a drop of water on the way in, so a pressure test that holds proves nothing about them. The shop pressure tests, finds nothing, tops up the vents, and comes back next season to the same air binding and the same black water.

The discriminator is not a pressure test. It is a makeup volume that has to be accounted for, and the routes that cannot account for it are the ones you should be looking at.

Before you change a fill setting or close a makeup valve

Do not raise a cold fill pressure until you have read the relief valve setting and the expansion tank precharge. Raising fill pressure narrows the headroom between cold static pressure and the relief setting, and a relief valve that lifts in an occupied mechanical room discharges water at boiler temperature into the room. Confirm the numbers first, on the tags, not from memory.

If you close the automatic makeup valve to watch a loop, the loop is now unprotected against water loss. Do it only with a functioning low water cutoff verified on the appliance, only with the system attended, and open the valve again before you leave the site. A boiler that runs low on water while somebody is off getting lunch is a far worse outcome than an unanswered diagnostic question.

Adjusting an expansion tank precharge means isolating the tank and draining its water side. That is stored energy in a pressure vessel, which is 29 CFR 1910.147, and the tank stays isolated and vented while the gauge goes on the air valve, not the other way round.

Read the loop temperature before opening any drain or vent. Above roughly 140 F, water leaving an opening flashes and scalds; isolate and let it cool, and put the discharge into a container placed where the stream cannot reach anybody.

Two families, and only one of them shows up on a meter

Sort every candidate route by a single question: does it lose water?

Routes that cost water. These appear on the makeup meter and can be found with a pressure test.

• Leaks at packing, seals, unions, buried or embedded lines, or a coil.
• A relief valve weeping on heat up, which is nearly always a failed or wrongly precharged expansion tank rather than a bad relief valve.
• An automatic air vent passing liquid instead of gas.
• Repeated draining for service that nobody logged.

Routes that cost no water at all. These are invisible to a pressure test and to a leak search, and they are where a healthy makeup number sends you.

• Any part of the system sitting below atmospheric pressure. An automatic float vent at negative gauge pressure is an air inlet, not an outlet. So is packing, so is a threaded joint that seals fine against outward pressure.
• Permeation through polymer. Oxygen diffuses through the wall of non barrier flexible tubing and through elastomeric hoses and expansion joints without any pressure difference driving it and without any water crossing the other way. Barrier rated tubing exists specifically to stop this.
• A direct air to water interface. A plain steel compression tank, or a bladder tank whose bladder has failed, puts atmospheric air in permanent contact with system water. Oxygen dissolves into it continuously, forever, at a rate set by the interface area and the temperature.

What each route cannot explain

This is the useful half. A route is eliminated when the thing it produces does not match what you are seeing.

• A leak cannot produce heavy corrosion product with a quiet makeup meter, because the oxygen arrives dissolved in the makeup water and the makeup water is the meter reading.
• Permeation cannot produce a large volume of free gas at a vent quickly. It is slow and steady, and it shows as corrosion product and gradually falling inhibitor rather than as a radiator that needs bleeding every fortnight.
• A failed bladder cannot hide from a gauge on the tank's air valve. Depress it: water means the bladder is gone.
• Sub atmospheric operation cannot occur on a system whose high point stays above atmospheric in every condition, and that is an arithmetic question you can settle in five minutes without touching anything.

The elimination on a four storey building

Hydronic heating loop, measured at 1,200 gallons. Complaint: the top floor emitters need bleeding every few weeks and the strainer loads with black fines. A makeup meter was fitted the previous year and reads 18 gallons for the year.

Start with the meter, because it eliminates a whole family at once. Eighteen gallons against 1,200 is 1.5% of the loop's volume replaced in a year. Cold makeup water carries dissolved oxygen on the order of 10 mg per litre, and 18 gallons is about 68 litres, so the year's makeup brought in roughly 0.68 g of oxygen. Oxygen converts iron to magnetite at roughly two and a half grams of iron per gram of oxygen, so the makeup for the entire year could produce on the order of 1.8 g of iron oxide. The strainer alone was collecting more than that every month.

Makeup borne oxygen is off by orders of magnitude, so every route that costs water is eliminated. There is no point pressure testing this building, and the pressure test the shop had already run and passed was not a false result, it was an answer to the wrong question.

Now the three routes that cost nothing. The tank was a bladder type and the gauge on its air valve read air, not water, so the interface route is out. The system was copper and steel throughout with no polymer tubing and no elastomeric expansion joints, so permeation is out. That leaves one candidate, and it is the one you can check with a tape measure.

The arithmetic that named it

Water column converts to pressure at about 0.433 psi per foot. The highest emitter on the top floor measured 42 ft above the pressure gauge at the boiler, so it takes 42 times 0.433, about 18.2 psi, simply to bring that point to zero gauge with the system static and cold. Add a margin of a few psi so vents stay positive and the water at the top does not flash, and the cold fill should be near 22 psi.

The fill valve was set at 12 psi, which is the common factory default and is a reasonable number for a two storey house. On this building it puts the top of the system at 12 minus 18.2, about 6 psi below atmospheric, whenever the system is cold and still. Every automatic vent up there is an inlet under those conditions.

There is a second mechanism on the same system and it acts at a different time, which is worth being able to separate. The expansion tank connected at the circulator's discharge rather than its suction. The tank connection is the point in a loop whose pressure does not change when the pump starts, so with that arrangement the pump's head is subtracted from the rest of the system when it runs instead of added to it. A 12 ft head circulator is about 5.2 psi, so the top of this system goes several psi further negative while the pump runs. That figure is an approximation that treats the high point as sitting close to the pump inlet in the friction chain; where the high point actually sits changes how much of the pump head lands on it, so use it to establish direction, not to quote a number to the owner.

The two mechanisms answer to opposite conditions. The static one is worst when the system is cold and off, because the water has contracted and pressure is at its lowest. The pump one is worst when the circulator is running. A building that shows both is drawing air across most of its operating cycle, which matches a complaint of bleeding every few weeks rather than every few years.

Why raising the fill is not a knob turn

The obvious fix is to raise the cold fill from 12 to about 22 psi. It is the right fix and it is three checks, not one.

Headroom to the relief. A 30 psi relief with a 22 psi cold fill leaves 8 psi for thermal expansion. Whether that is enough depends on the tank, and if it is not, the relief lifts on heat up, dumps water, and the makeup valve replaces it with fresh oxygenated water. You would have converted a no water ingress problem into a water losing one and made the corrosion worse.

Precharge. The tank's air precharge is set to match the cold fill pressure at the tank's location, with the tank isolated and its water side drained. Raise the fill without raising the precharge and the tank is partly waterlogged at fill, so it has less acceptance volume and system pressure climbs faster on heat up, which walks you back into the relief valve.

Acceptance volume. Narrowing the operating pressure band reduces how much expansion a given tank can absorb. Recheck the tank sizing against the new band and the system volume rather than assuming the existing tank still suits.

And relocate the tank connection to the circulator's suction side, or move the circulator, so pump head adds to the system instead of subtracting from it. That change costs a fitting and an afternoon and it removes the second mechanism permanently.

What would point somewhere else

• A makeup meter that does move. If the year's makeup had been several system volumes rather than 1.5%, the elimination above reverses entirely and the job is a leak search, starting with the relief valve discharge and the expansion tank.
• Radiant floor tubing, or hoses and flexible connectors. Permeation becomes a live candidate the moment there is polymer in the wetted path, and non barrier tubing in a heating loop is a known and unfixable oxygen source short of replacing it or separating it behind a heat exchanger.
• A chilled water loop rather than a heating one. The static and pump mechanisms are identical, but a chilled loop never drives its dissolved gas out at a vent, so it gives you no air binding complaint at all. The same ingress shows up only as corrosion product and falling inhibitor, which is why chilled loops get diagnosed years later than heating loops with the same fault.
• A system with no high point vent complaint but heavy sludge. Air binding is a symptom of free gas collecting. Its absence does not mean no oxygen is entering, and the sludge is the more reliable witness.

References

• 29 CFR 1910.147, The Control of Hazardous Energy, for isolating and venting an expansion tank before working on its precharge
• Expansion tank and relief valve manufacturer documentation for precharge setting, acceptance volume and relief setpoint
• ASHRAE Handbook, HVAC Systems and Equipment, for expansion tank sizing and the point of no pressure change in a closed loop
• See related: What Dissolved Oxygen Does Inside a Sealed System, which owns the oxygen mass arithmetic used in the elimination here