FDD vs predictive maintenance: what is the difference for facility teams?
Milvian AI teamFacilities and infrastructure intelligence
Fault detection and diagnostics (FDD) is reactive: it detects a fault that has already happened, diagnoses the cause and gets it fixed. Predictive maintenance is proactive: it catches early signs of failure, such as vibration, a changing sound signature or energy drift, so maintenance happens before the fault.
Why do the two terms get confused?
Vendors often use fault detection and diagnostics and predictive maintenance interchangeably. They are not the same thing. They answer different questions, at different moments, from different data. Mixing them up leads to buying the wrong capability, or expecting one to do the job of the other.
The simplest way to separate them is time. FDD works after the fault: something has gone wrong and the job is to find it, explain it and get it fixed fast. Predictive maintenance works before the fault: nothing has failed yet, and the job is to spot the early signs and schedule the work while it is still cheap.
How do they compare?
| FDD | Predictive maintenance | |
|---|---|---|
| When it acts. | After the fault. | Before the fault. |
| Question it answers. | What has gone wrong, and why? | What is about to go wrong? |
| Typical inputs. | BMS points, setpoints, status, meter data. | Vibration, sound signature, energy drift, runtime. |
| Example. | An AHU fan has stopped. Cause: failed rotor shaft. | A compressor shows new vibration and rising energy use. |
| Output. | Diagnosis and a fix routed for repair. | Maintenance scheduled ahead of failure. |
| How you know it worked. | Airflow or output is confirmed back in range. | The failure does not happen; the asset stays in range. |
How does FDD work?
FDD compares what equipment is doing with what it should be doing. Rules and models read sensor values, commands and status from the BMS and from meters, and flag a fault when they disagree: a fan commanded on with no airflow, a valve commanded closed while the coil still heats, a supply air temperature that will not reach its setpoint, a damper that never moves.
Detection alone is not enough. A good FDD system diagnoses the likely cause in plain language, ranks the fault by cost and comfort impact, and sends it to the team that will fix it. Take an air handler whose blower fan stops. FDD flags the stop, diagnoses a failed rotor shaft, routes the repair into the CMMS, and then confirms airflow is back once the work is done. Faults found, diagnosed and fixed.
FDD also catches faults that never trip an alarm. Simultaneous heating and cooling, an economizer stuck at minimum, or a schedule that runs all weekend may not stop anything, but they waste energy every hour they persist.
How does predictive maintenance work?
Predictive maintenance looks for change in how an asset behaves before it fails. The early signs are physical. A compressor starts to vibrate differently. A pump's sound signature changes. A motor draws more energy for the same work. Each of these can appear well before a breakdown.
The method is to learn each asset's normal pattern and watch for drift away from it. Vibration and acoustic sensors capture the mechanical signs. Energy metering captures efficiency loss. When drift crosses a threshold, maintenance is scheduled before the asset fails, at a time that suits the operation. Stops failures before they happen.
Predictive maintenance is related to condition-based maintenance, which acts on the measured condition of an asset rather than a fixed calendar. Both aim to replace routine time-based servicing with work driven by evidence.
What mistakes do buyers make?
The first mistake is treating alerts as the outcome. An FDD tool that produces hundreds of fault notices a week, with no ranking and no route to a fix, adds work instead of removing it. Ask how faults are prioritized, how they reach the people who fix them, and how the system confirms the fix worked.
The second is expecting prediction without the right data. It needs signals that move ahead of a breakdown: vibration, acoustic and energy data at the asset. A BMS status point that says on or off cannot predict anything. If the critical assets are not instrumented, start with FDD and add sensing to the assets that matter most.
The third is stopping at the ticket. A closed work order is not proof that the problem is solved. The loop only closes when the meter or sensor shows the asset back in range, and the energy it uses back to its baseline.
Which one do you need?
Most facilities need both. FDD gives the fastest return in buildings with an existing BMS, because the data is already there and many faults are already wasting energy. Predictive maintenance pays most on critical rotating equipment, such as chillers, compressors, pumps and large fans, where an unplanned failure is expensive or disruptive.
Neither replaces the CMMS. FDD and predictive maintenance decide what needs doing and why. The CMMS manages technicians, parts and work orders. The two should be integrated, so a finding becomes a ticket with the diagnosis attached, and the closure comes back to be verified.
