Fault detection on a BMS - ten rules that find the most common waste
Most of the waste in an existing BMS is not hidden. It is sitting in the trend logs - a valve open when it should be shut, a pump at full speed in the middle of the night, two coils fighting each other in the same air handling unit. Fault detection and diagnostics (FDD) is simply the discipline of looking for those patterns systematically, instead of hoping someone notices.
This guide sets out ten classic rule-based checks that integrators and engineers use, the points each one needs, the traps that cause false alarms, and what the published evidence says about the results.
What FDD is
At its simplest, FDD applies rules to trend data: "if this and this are true for longer than that, flag it". Rules can be run by hand on exported trends during a service visit, or continuously by analytics software connected to the BMS.
The difference matters less than it seems. The value comes from three things: having the right points trended, applying sensible rules, and - most importantly - someone acting on what the rules find. Software makes it continuous and scalable; it does not replace the engineer who decides what to do.
The ten classic rules
1. Simultaneous heating and cooling
What it looks for: heating and cooling valves (or a heating valve and a cooling coil, or reheat and chilled air) both open in the same unit or zone at the same time.
Points needed: heating and cooling valve positions or commands for the same unit; supply and space temperatures help to confirm.
This is often the single largest finding on an existing system, especially in the mixed-mode weeks of spring and autumn.
2. Passing valves
What it looks for: a temperature change across a coil when its valve is commanded shut - a sign that the valve is not sealing, or a bypass is open.
Points needed: valve command, on- and off-coil air temperatures (or flow and return water temperatures).
Allow for a short delay after the valve closes, and for heat picked up from the fan, before flagging.
3. Hunting loops
What it looks for: a control output that oscillates repeatedly - a valve or damper swinging open and shut, or a temperature cycling around its setpoint.
Points needed: the loop's output and its controlled variable, trended at a short interval (a few seconds to a minute, depending on the loop).
Hunting wears out actuators and wastes energy. The usual causes are poor tuning, an oversized valve or a sensor in the wrong place.
4. Plant running out of hours
What it looks for: plant running when the building is meant to be unoccupied.
Points needed: plant run status, the occupancy schedule and any holiday calendar.
Some out-of-hours running is legitimate - optimum start, frost protection, night purge. The rule needs to know about these, or it will flag them.
5. Fixed setpoints where a reset is specified
What it looks for: a supply temperature or pressure that never moves, where the strategy says it should reset with load or outside conditions.
Points needed: the setpoint being reset, and the variable it should be reset from (outside air temperature, valve positions, zone demand).
A reset that has been overridden or disabled is easy to miss because nothing visibly fails - the plant simply works harder than it needs to.
6. Stuck dampers
What it looks for: a damper that does not move when commanded, or a mixed air temperature that does not match the damper position.
Points needed: damper command and, ideally, position feedback; outside, return and mixed air temperatures.
A stuck outside air damper can mean too much fresh air in winter, too little ventilation in summer, or free cooling that never happens.
7. Sensors flatlined or out of range
What it looks for: a sensor whose value does not change at all for a long period, or reads outside a physically sensible range.
Points needed: the sensor itself, trended at a reasonable interval.
Flatlined sensors are often failed or disconnected. Out-of-range values can be wiring faults, wrong scaling or a sensor of the wrong type. Every control loop that depends on that sensor is affected.
8. Low chilled water delta-T
What it looks for: a difference between chilled water flow and return temperatures well below the design value.
Points needed: chilled water flow and return temperatures, and ideally flow rate and chiller status.
Low delta-T means more water is pumped for the same cooling, and chillers can be forced to run more stages than needed. Passing valves, coils with no valve control and incorrect bypasses are common causes.
9. Boilers short-cycling
What it looks for: a boiler or heat pump starting and stopping many times an hour.
Points needed: run status (or firing signal) for each heat source, trended at a short interval.
Short-cycling wastes energy and wears plant. Causes include oversized plant at low load, tight control differentials and poor sequencing between multiple heat sources.
10. Pumps and fans at 100% with the drive in hand
What it looks for: a variable speed drive running at full speed for long periods, especially when the drive is in hand or local control rather than following the BMS.
Points needed: drive speed or output, the drive's hand/auto status, and the pressure or demand it should be controlling.
A drive left in hand at full speed after a fault or a commissioning test gives away much of the saving it was installed to provide.
False positives
The fastest way to make people ignore FDD is to fill their inbox with findings that are not real. Common causes of false positives:
- No delay or persistence time - flagging a condition the instant it appears, instead of after it has persisted.
- Ignoring legitimate modes - frost protection, optimum start, purge and commissioning tests all look like faults to a naive rule.
- Bad sensors - a rule that relies on a failed sensor will produce nonsense. Check sensor health first (rule 7).
- Wrong trend intervals - a 15-minute log will not show hunting; a one-second log of everything wastes storage.
- Not knowing the strategy - a rule should check the plant against what it is meant to do, not against a generic idea of good practice.
Start with a few rules, tune them on one building and add more as trust builds. A short list of findings that are real is worth far more than a long list nobody believes.
Acting on findings - the loop that delivers the savings
Finding a fault saves nothing. The savings come from the loop that follows:
- Find - the rules flag a fault.
- Confirm - an engineer looks at the trends and, if needed, the plant.
- Fix - on the next visit, or remotely where that is safe and agreed.
- Prove - the trend after the fix shows the fault has gone, and metering shows the effect.
- Record - the finding, fix and evidence go into the service report.
This is the strongest argument for remote monitoring in a service contract: the monitoring finds it, and the visit fixes it. Visits become targeted - "we are coming to fix these twelve things" - rather than a general walk-round.
What the published evidence says
The best-known published evidence comes from the US. The Smart Energy Analytics Campaign, run by the US Department of Energy with Lawrence Berkeley National Laboratory, followed organisations using energy information systems and FDD software. After two years, the US DOE reported median savings of about 9% for FDD users, against about 3-4% for meter-level energy information systems, and a simple payback of one to two years. The reports also noted that savings depended on people acting on the findings.
Other US studies point the same way. Lawrence Berkeley National Laboratory analysed FDD data from 317 buildings, covering 3,660 air handling units and nearly 54,000 terminal units (Crowe and others, 2023). On a typical day, around 40% of the air handling units and 30% of the terminal units had at least one fault reported. The most common air handling unit fault - supply air temperature not meeting its setpoint - appeared on 55% of units, and zone temperature sensor faults on 46% of terminal units. The Pacific Northwest National Laboratory, working from BMS trend data on 151 buildings (Katipamula and others, 2021), found operational problems in almost every building, with a median savings potential of 12% and a range from zero to 50%.
| AHUs - supply air temperature not tracking setpoint | 55% of AHUs |
|---|---|
| Terminal units - zone temperature sensor fault | 46% |
| Terminal units - abnormal zone temperature | 46% |
| AHUs with at least one fault on any given day | 40% |
| Terminal units with at least one fault on any given day | 30% |
Common and persistent AHU faults: supply air temperature and static pressure setpoints, the economiser sequence, the supply air sensor, coil valves and airflow sensors.
These are median results from particular groups of North American buildings and organisations, not a prediction for any one building. Results elsewhere will depend on the building, how well it was running before, and how consistently the findings are acted on. Use them to explain why FDD is worth doing, never as a promise.
Getting started
You do not need a full analytics platform to start. Pick one building, make sure the points listed above are trended, and work through the ten rules by hand on a week of data before the next visit. That alone usually gives a useful findings list.
For more on putting this into a service programme, see The BMS service visit that pays for itself and The BMS health check. Many of the fixes are the measures described in Where BMS tuning saves energy, and Proving savings shows how to demonstrate the result. If the faults trace back to how the system was set up, Commissioning a BMS covers the original sequence, and Seasonal changeover covers the faults that appear twice a year.
Related tools and training: the out-of-hours running cost, VSD savings and PID tools are in tecTools, and the Hydronics and valves and AHU commissioning courses are in tecTraining.
Published evidence mentioned (all from the US): US DOE and Lawrence Berkeley National Laboratory, Smart Energy Analytics Campaign reports (2019-2020); LBNL fault prevalence study (Crowe et al. 2023); PNNL operational problems study (Katipamula et al. 2021). Results vary by building and are not a guarantee of savings.