NFPA 70E draws the arc flash boundary where incident energy reaches 1.2 cal/cm², about 5 J/cm², the level at which a second-degree burn on bare skin becomes likely. The distance is calculated per board from fault current and clearing time, so a small remote board can carry a wider boundary than the main switchboard.
An action alarm set at the specification limit leaves no margin for the sensor's uncertainty, so the room may already be outside when it rings. Borrow the guard band from calibration practice: with a 60% RH limit and ±2% RH uncertainty, set the action alarm at 58%, add an alert before it, and record the reasoning with the calibration data.
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.
Below 0 °C, moisture in air can be stated as a dew point over supercooled water or a frost point over ice, and the frost point is the warmer number. At a frost point of −40 °C the dew point is about −43.7 °C. A dry-room specification should say which one it means.
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.
A vague site note gets no reply. One that gets acted on says where and when, what was seen (measured where possible), what the drawing or specification requires, what happens if it is left, and what action is proposed, by whom and by when. It takes ten minutes and survives being forwarded.
Some reheat is the price of dehumidification; the rest is waste. Five BMS trends separate the two: cooling and reheat valve positions together, supply dew point against what the rooms need, overlapping setpoints, temperature change across a coil with its valve shut, and the same pattern at night and weekends.
FMEA is one of nine tools listed in ICH Q9(R1). The process around it has five parts: initiate with a defined risk question, assess (identify hazards, analyse, evaluate), control (reduce, then accept), communicate the output, and review when events or new knowledge arrive. A first page with the risk question repairs most thin files.
Bracketing, as ICH Q1D (2002) defines it, tests only the extremes of a design factor such as strength or container size at all time points and assumes they represent the levels between. Three strengths in two bottle sizes still leave four of six combinations on test, and any reduced design should be justified.
Once the last ceiling panel is in, the services above it are hard to inspect. Before close-up, check that hangers are complete, insulation and vapour seal are continuous, dampers and valves are tagged and reachable, penetrations are sealed, and test ports are fitted with no debris left. Photograph each bay for the handover package.
ICH Q2(R1) said accuracy should be assessed using a minimum of 9 determinations over at least 3 concentration levels. Q2(R2), adopted in November 2023, asks for an appropriate number and keeps 3 concentrations with 3 replicates as its example. The minimum of 9 remains under repeatability, and the confidence interval for accuracy should be compatible with the acceptance criterion unless otherwise justified.