A filter changed on night shift without a leak test turns maintenance into a deviation. GMP maintenance works from criticality bands built on the asset register, planned and corrective work in one system, risk-based requalification and calibration, and 24/7 response across Southeast Asia under a maintenance agreement.
A sterile room is positive to protect the product; a high-potency OSD room must keep powder in. A cascade copied from a sterile template sends API dust into the corridor at every door opening. Check every door arrow, choose sink or bubble airlocks on purpose and extract through bag-in/bag-out HEPA housings.
Filter withdrawal, coil pull space and fan replacement all need clearance that nobody draws by default. Dimensioning these zones on the AHU general arrangement drawing and carrying them into the BIM model keeps other trades out of them, and avoids a shutdown the first time a coil has to come out.
EU GMP Annex 1 (2022) sets maximum requalification intervals: 6 months for Grade A and B cleanrooms and 12 months for Grade C and D. Classification, HEPA leak tests, airflow volume, pressure differentials and air velocity, or recovery where airflow is not unidirectional, are retested. A risk assessment can shorten these intervals, never extend them.
EN 1886 class L2 allows an AHU casing 0.44 l/s per m² of leakage at −400 Pa. On a witnessed FAT, casing leakage, fan airflow at design static pressure, filter bypass and the control panel are checked against the approved specification, and both engineers sign the sheet that goes into the equipment file.
A new AHU started during construction loads its coil, drain pan and final filters with site dust that does not vacuum off wet fins. Clean the casing before power-on, seal duct openings, run on temporary pre-filters, fit final filters after the final clean and record it for IQ.
ISO 14644-3 allows installed filter leak scans by aerosol photometer or discrete particle counter. The photometer is quick but needs roughly 10 to 100 mg/m³ upstream; the counter works at far lower loads. For ULPA and filters you want kept free of oil, choose the counter method at design stage.
A photometer leak scan passes a filter when downstream readings stay below 0.01% of the measured upstream concentration. At 20 µg/L upstream the threshold is 0.002 µg/L; a thin challenge pushes it toward the instrument's zero and a real leak reads clean. Reports should state the upstream figure per filter and where it was sampled.
The part that takes a graded room out of spec is usually cheap: a transmitter, a belt, an actuator. Sort spares by what a failure does to the room and how long a replacement takes to arrive: hold on site, hold with a supplier, or order on demand. Review after every breakdown.
A Grade A particle counter samples one point and says nothing about the path air takes to the product. First air must reach an open container without passing over gloves or components. An in-operation smoke study that includes interventions is the evidence, and its findings belong in the Contamination Control Strategy.
Under EN 779, F9 was a filter class: 95% or more at 0.4 µm. Under EN 1886, F9 is a casing class: at most 0.5% of the airflow bypassing the filter frames. Specify filter stages to ISO 16890 or EN 1822 and the casing bypass class on its own line.