ICH Q10 expects changes to be evaluated after implementation. Closing a change the day the part is installed skips that step. Define the effectiveness check when the change is approved: the data that shows the objective was met, a period long enough to include the original conditions, and the result that closes it.
"OK" on a FAT sheet records a judgement. "118 Pa, acceptance ≤ 150 Pa", with the gauge ID and calibration date, records a measurement that IQ, maintenance and an inspector can use years later, and shows when a pass was close. It is how the Excellence value looks on a test bench.
Set a cleanroom pressure alarm delay from OQ recordings: log each door opening at one-second resolution, use the longest normal recovery plus margin as the floor and the risk assessment as the ceiling, record the justification per room pair, and add a separate door-held-open alarm.
A failed acceptance criterion and a protocol step not executed as written are different OQ deviations. The first needs root cause, impact assessment, a fix and a repeat test; the second needs an assessment of whether the result still stands. Changing a criterion after a failure invites audit questions.
A 100 cm² door gap at 15 Pa passes about 110 m³/h, but only doors on the suite boundary cost fresh air. Internal door leakage returns through the neighbouring room's grille. For twenty doors with four on the boundary, the leakage to make up is 440 m³/h rather than 2,200 m³/h of wet tropical air.
A full-size 592 × 592 mm bag filter is commonly rated at about 3,400 m³/h. A 20,000 m³/h AHU needs six frames; a casing with room for four runs each at 47% over, with higher pressure drop, possibly lower efficiency and fluttering pockets. The frame count belongs in the specification and the FAT.
Operators run the plant from BMS graphics, so errors there become operating errors. Check each screen for five things: every value has a unit and a tag, the setpoint sits beside the actual value, alarm colours follow one convention, stale values look different from live ones, and room names match the room codes.
A room pressure loop tuned with every door shut hunts once production starts. Balance the airflows with the loop in manual, stroke the damper, tune from the reference room outward, then open and close doors and record the swing and the settling time. Those records are what the alarm delays should rest on.
Sensible heat in watts equals 0.335 × airflow in m³/h × temperature difference in kelvin, so 10,000 m³/h across 10 K carries 33.5 kW. The constant comes from air density and specific heat. Correct it for density at altitude, and remember it leaves out the latent load.
Adjusting each diffuser to its design flow in turn goes round the room without settling, because terminals on a branch share one duct pressure. Proportional balancing sets every terminal to the same ratio of design, with the lowest-ratio terminal left fully open, and only then brings the fan to the design total.
Fan curves are measured with a uniform inlet and a straight outlet duct. An elbow on the outlet or a wall near the inlet makes the fan deliver less pressure at the same speed, a loss the ordinary fitting coefficient does not include. AMCA Publication 201 gives system effect factors to add at selection, or keep the straight lengths.
EN 1886 class L1 limits casing leakage to 0.15 l/s per m² at −400 Pa; L2 allows about three times more. As filters load and casing pressure rises, leakage grows, room offsets shrink and pressure control hunts. Specify L1, filter bypass F9 and a witnessed leakage test at FAT.