The channel we could not clean, and the redesign that followed

A geometry that trapped what it was supposed to flush.

The channel we could not clean was a short blind pocket sitting behind a valve seat, a feature that had looked like a harmless dead end on the drawing and turned out, once liquid actually flowed past it, to be a corner no rinse cycle at any flow rate or duration could fully reach.

A shoulder that left a recess behind it

The pocket existed because the valve seat needed a shoulder to close against, and the simplest way to cut that shoulder into the block left a small blind recess directly behind it. On paper it looked like the many other minor pockets in a typical manifold and was never flagged during design review. Once the block was tested with liquid, the main flow swept past the valve seat at speed on its way to the outlet, but almost none of it entered the recess, since entering meant turning sharply against the current. A rinse cycle run far longer than any other channel in the block needed still left a detectable trace of the previous liquid in that one recess.

Pushing a pipe cleaner through a bent drinking straw clears the straight sections easily, but the wire bunches and buckles at a sharp bend, and the inside of that bend often comes away barely touched even when the rest of the straw looks spotless. The valve pocket was the manifold's version of that bend: every other flushing action in the design handled its channels without difficulty, and in this one place the geometry itself worked against being cleaned. Pushing the pipe cleaner harder only makes it buckle sooner, and running the rinse faster did much the same to the manifold, since a quicker main stream skims past the mouth of the recess with even less inclination to turn into it.

Dead volume and an uncleanable corner are one fault

A flush is the same flow pattern as normal operation, repeated with a different liquid, so a pocket that receives little of the working flow receives just as little of the rinse. A feature identified as dead volume for contamination purposes is, by the same mechanism, a feature that resists cleaning once contamination has happened. The two problems were one geometric fact showing up twice, once as a slow contamination trail and once as a residue that extra rinsing could not shift.

That is why longer rinse times, higher flush pressures and more aggressive cleaning agents, all tried before the redesign, each improved the result slightly without solving it. A pocket that exchanges its contents far more slowly than the main channel needs a correspondingly longer flush, and the required time becomes impractical well before the pocket is clean. Partial fixes of this kind tempt a team into believing a problem has been managed, while leaving the pocket exactly where it was.

Redrawing the seat flush with the wall

The eventual fix changed the geometry. The valve seat was redrawn so its shoulder sat flush with the main channel wall, which meant it no longer needed a blind cavity behind it, and the volume that had been unreachable became part of the swept channel like everywhere else in the block. The valve sealed exactly as well as before, and the flush procedure went back to normal, since there was no longer a hidden pocket to need special treatment. Fixes like this tend to look disappointingly simple next to the effort the workaround stage had already absorbed.

The lesson carried into how every later manifold drawing in that project was reviewed. A reviewer checking a design for dead volume stopped asking whether a feature could be flushed given enough time and pressure, and asked whether it received a genuine share of the main flow during ordinary operation, since only that second question reliably predicts whether cleaning will succeed.

My key error with this

I designed that part holder around the coolant, which was the obvious fluid to think about, so there were paths for it to run off and nothing anywhere that would let it pool against the work. What I had not thought about was everything the coolant carries with it once cutting starts, because fine swarf and the low viscosity sludge that comes off a machining operation do not behave like clean coolant, and they settled into a pocket in the holder that the flow ran straight over rather than through. The pocket filled slowly enough that nobody noticed until parts started coming out of tolerance for reasons that made no sense on the machine, and cleaning it out properly meant taking the fixture off and stripping it. Adding a drip tray under the holder and a channel to carry the residue to a single collection point solved it completely and should have been there from the first version. What replaced the belief is that designing for a fluid means designing for what the fluid is carrying by the end of the job, and that the question is never where the liquid goes, it is where the solids it drops end up.