How to Tell an Internal Leak From an External One

Why this matters

An external leak takes fluid out of the machine. An internal leak takes capacity out of it and leaves every drop of fluid exactly where it was. That single difference splits every "it is slow" and "it will not hold" complaint into two classes with opposite search methods, and a tech who has not made the split spends the visit doing the wrong one. Crawling under a machine with a flashlight finds external leaks and will never find an internal one, no matter how thorough you are, because there is nothing to find.

Make the classification first, in about fifteen minutes, before any wrench comes out.

Before anything: the machine is running for most of this

Two of the steps below need the machine under load, so the safety scope is the whole procedure rather than one step. Readings are taken from outside the motion path, with guards in place, on a machine under its own controls. Nothing is touched by hand: line and cylinder surfaces at operating temperature burn skin on contact. Never stand in line with a hose end or a fitting, and never run a hand along a line to hunt for a leak, because a pinhole stream injects through skin and that is a surgical emergency rather than a cut, covered in its own card.

Any step that opens a joint happens with the machine shut down, the circuit relieved to zero through its own controls rather than by cracking a fitting, and any accumulator discharged under the plant's energy control procedure, which is required for stored energy in general industry by 29 CFR 1910.147. An accumulator holds full working pressure with the pump locked out.

Step 1: Read the reservoir level, then apply its three corrections

The level is the only measurement that separates the classes directly, and it is also the measurement people misread most often, because three things move it that have nothing to do with a leak.

• Rod position. An extended cylinder is holding fluid that would otherwise be in the tank, and a machine parked with rods out reads low. Compare only at a defined rod position, and say which one.
• Temperature. The charge expands as it warms. A level taken hot and compared to a mark set cold reads high, which can mask a real loss. Compare at the same temperature, which in practice means at shift start before the warm-up run.
• Accumulator state. A charged accumulator is holding fluid out of the tank. Compare with it in the same state both times, and if you cannot, say so.

Write the raw reading, write each correction, and write the corrected result. A single "it looks about right" here is what sends a shop under the machine for a day.

Step 2: Read the top-up history before you generate new data

A machine that has been topped up on a schedule has been losing fluid for as long as that schedule has existed, and somebody already measured it. Ask for the log, or ask the operator how often they add and how much. Months of evidence sit in that answer, and it costs one question.

No top-ups and a stable corrected level is a strong negative. Regular top-ups with no visible wet spot is not a contradiction, and the third step covers why.

Step 3: Know the two external leaks that leave no puddle

Before you conclude internal from a missing puddle, rule these out, because both are external losses that go somewhere tidy:

• An oil-to-water cooler with a failed core puts oil into the cooling water. Check the water side for oil film or an emulsion, and check whether anyone has been draining a milky cooling circuit.
• A leak into another enclosed volume, such as a gearcase, a wet-sump housing or a machine base that drains internally. The tell is a level rising where it should not, in a compartment nobody thought to open.

Both of these read as "no external leak" on a walk-around and are external in every way that matters, including the fact that adding filtration and changing seals will not stop them.

Step 4: Read the energy balance, which is where an internal leak shows up

An internal leak is pressure falling without an actuator moving, and that becomes heat in the oil. So a machine losing capacity internally runs hotter than its own baseline, and a machine losing fluid externally does not, because the fluid takes its energy out of the system with it.

Use the reservoir rise above ambient against the machine's commissioning baseline, and treat a ratio within about 1.2 of baseline as inside ordinary spread. The sibling cards on heat as a symptom and on finding the heat source own that gate and the component-level survey. This step only asks whether the balance moved.

Step 5: Separate pressure-dependent slowness from flow-independent slowness

An internal leak passes more fluid the harder you push on it, so it steals more at high load than at low load. A restriction, a wrong drive speed or a mis-set flow control does not care about the load in the same way.

Time the same actuator through the same stroke twice, once unloaded and once at working load, three runs each, and take the median of each set. Normal unloaded and slow loaded is pressure-dependent: an internal leak. Slow at both is a flow or restriction problem and this article stops being the right one.

The error on those timings, with its basis and character. Hand timing carries an independent random spread of a few tenths of a second, and it lands twice in each run, on the start and on the stop. Independent spreads combine in quadrature, so the two multiply by the square root of two rather than adding. Taking the median of three runs shrinks it further. That residual is what tells you whether a difference between two timings is a finding or a stopwatch.

Step 6: Find the component, by section rather than by inspection

Internal leakage is found by blocking a section and asking whether the symptom survives. The pump gets one test the rest of the circuit does not: its case drain flow, timed into a container, against the builder's published maximum. That number belongs to the machine builder and is worth asking for before the visit.

Route the case drain hose into a secured graduated container before the machine starts, with gloves and eye protection on, and never hold a container in a fluid stream or disconnect a case drain line on a running machine. Time the fill from outside the motion path and shut the machine down before the line is put back.

Worked example: two machines, one procedure, opposite answers

Both machines came in with the same complaint, an actuator that used to complete its stroke in about 8.0 seconds and now takes about 11.5 seconds, which is 11.5 divided by 8.0, so 44 percent longer.

Machine one, the level. Raw sight glass reading, taken hot at the end of a shift, sits 1.0 inch below the mark. Corrections: the mark was set with all rods retracted and two rods are currently extended, which accounts for 0.8 inch, so 1.0 minus 0.8 leaves 0.2 inch. The remaining comparison is hot against a cold mark, so it was retaken cold at shift start with rods retracted, and the residual was 0.1 inch, which is inside the resolution of that sight glass. Corrected result: no fluid has left the system. The top-up log shows no additions in 14 months, which agrees rather than merely failing to disagree.

Machine one, the balance. Ambient 72 degrees F, reservoir 140 degrees F, rise 68 degrees F. Commissioning baseline rise 45 degrees F. Ratio 68 divided by 45 is 1.51, above the 1.2 band, so generation moved. Cooler drop checked at 13 degrees F against a 12 degree F baseline, so the cooler is rejecting more rather than less and this is not a cooling failure.

Machine one, the load test. Unloaded stroke, median of three: 8.2 seconds. Loaded stroke, median of three: 11.5 seconds. The unloaded figure sits within the stopwatch residual of the 8.0 second commissioning time. The loaded figure does not. Pressure-dependent: internal.

Machine one, the component. Case drain flow timed into a secured container came back at 3.1 times the commissioning figure and above the builder's published maximum. Deadheaded against the relief, the circuit reached about 90 percent of its relief setting where commissioning cracked the relief, which is the same finding read a second way. The pump's rotating group is passing fluid internally, and the fluid sample taken earlier explains why the clearances opened.

Machine two, the level. Raw reading 1.4 inches below the mark, taken cold at shift start with rods retracted, so the rod correction is zero and the temperature correction is zero. Corrected result: 1.4 inches of fluid has left the system. The operator adds fluid about every third week.

Machine two, the balance. Ambient 70 degrees F, reservoir 116 degrees F, rise 46 degrees F, against the same 45 degree F baseline. Ratio 1.02, inside the 1.2 band. Nothing is being turned into heat that was not being turned into heat before.

Machine two, the puddle that was not there. No wet spot anywhere on the walk-around, which is what had the shop convinced it was internal. Step 3 caught it: the water side of the oil-to-water cooler was carrying an oil film, and maintenance had been draining a milky cooling circuit for two months without connecting the two facts. External, through the cooler core.

Why machine two was slow. With a level that low and rods cycling, the pump was pulling air on the fastest part of the stroke. That is a suction-side problem created by the leak, not an internal leakage problem, and it resolved when the level came back after the core was replaced.

The failure mode both ways. Machine one would have been sent under the frame with a flashlight and a can of leak dye, and nothing would have been found because nothing was there. Machine two would have had its pump condemned on a slow cycle time, and it would have come back slow, because the new pump would have been pulling the same air from the same low level through the same failed core.

How to verify you classified it right

• State the corrected level, not the raw one, in the write-up. A raw reading with no correction line is not a measurement anyone can check, including you in a month.
• Confirm the class has two agreeing signals. Internal should show both a stable corrected level and a raised rise above ambient. External should show a falling level and an unchanged rise. One signal on its own is a hypothesis.
• Check your comparator was taken the same way. If the baseline rise was recorded at a different duty cycle, or the level mark was set at a rod position nobody wrote down, say so in the same sentence as the comparison rather than presenting the ratio as clean.
• Re-time the actuator after the repair, loaded and unloaded, median of three. If only the loaded time recovers, the internal path was the whole story. If neither recovers, the classification was wrong and step 5 is where to go back to.

References

• Machine builder documentation for the pump's maximum allowable case drain flow, the commissioning cycle times, the relief setting and the commissioning oil temperature rise
• 29 CFR 1910.147, for control of hazardous energy including stored hydraulic energy, in general industry, before any joint is opened or any line is disconnected
• Safety data sheet for the specific hydraulic fluid, governing skin and eye contact, spill response, and disposal of fluid drained during a case drain test
• See related: Why Heat in a Hydraulic System Is a Symptom Rather Than a Fault; What Cylinder Drift Tells You About Which Seal Gave Up; What Fluid Injection Injury Is and Why It Is a Surgical Emergency