Vibration analysis measures the vibration of rotating machinery and splits it into frequency components to find faults before a breakdown. Unbalance, misalignment, looseness and bearing damage each leave a distinct signature, so a well-placed sensor and a tracked trend can reveal a fault weeks early. This article covers measurement; for maintenance strategy, see our predictive maintenance guide.
Why vibration is the earliest warning you get
Bearings and couplings rarely fail without notice. Reliability engineers describe the process with the P-F curve: P is the point at which a developing failure first becomes detectable, and F is the point at which the asset can no longer do its job. The interval between them is the window you have for a planned repair.
Different symptoms appear at different points along that curve. A change in vibration is usually among the first; a rise in temperature tends to follow, and audible noise, smell or visible damage come last. By the time a temperature probe or an operator's ear raises the alarm, the fault is well advanced. Monitoring vibration moves detection earlier and gives the maintenance team time to order the spare part and schedule the stop for a quiet shift.
Where vibration monitoring sits in a plant's digitalisation sequence is shown in our Industry 4.0 roadmap for SMEs: it is the most mature technique of the prediction stage, which follows once data collection and visibility are in place.
What an undetected fault really costs
Unplanned downtime is expensive, and much of the cost hides in overtime, expedited freight, scrap and missed delivery dates.
Siemens estimates that the world's 500 largest companies lose almost $1.4 trillion a year to unplanned downtime, equivalent to 11% of their revenues. (Siemens, The True Cost of Downtime 2024)
The same report finds that large plants now average 25 downtime incidents a month and lose 27 hours a month to unplanned stoppages. Those figures are lower than in 2019, but a single critical machine can still hold up an entire line. Another finding matters here: nine out of ten respondents in the Siemens survey already do some form of condition monitoring. Vibration monitoring is no longer an experiment; it is becoming routine plant practice.
The pay-off has been measured too. Independent surveys cited in the US Department of Energy's Operations & Maintenance Best Practices Guide, Chapter 5 report that a functional predictive maintenance programme cuts downtime by 35–45% and maintenance costs by 25–30% on industrial average, and the guide puts savings over preventive maintenance alone at 8–12%. These are industry averages; what you achieve depends on your machine fleet, your current maintenance regime and the quality of your data.
Vibration analysis: which machine fault shows up at which frequency?
In the time domain a vibration signal looks like noise. An FFT (Fast Fourier Transform) separates it into frequency components, and each peak in the resulting spectrum points to a physical cause. Frequencies are usually read as multiples of shaft speed: 1X is the running speed itself (25 Hz for a shaft turning at 1,500 rpm).
| Fault | Typical spectral signature | Measurement direction |
|---|---|---|
| Unbalance | Dominant 1X peak, radial | Horizontal and vertical |
| Misalignment (coupling) | High 1X and 2X with a strong axial component | Axial |
| Mechanical looseness | Many harmonics of 1X (2X, 3X, 4X…) | Radial |
| Rolling-element bearing damage | Bearing defect frequencies (BPFO, BPFI, BSF, FTF) and harmonics | Radial, high frequency |
| Gear damage | Gear mesh frequency with sidebands | Radial / axial |
| Electrical faults (motor) | Twice line frequency (100 Hz on a 50 Hz supply) | Radial |
Bearings deserve particular attention. Damage that starts on the outer race (BPFO), inner race (BPFI), rolling element (BSF) or cage (FTF) appears at specific frequencies calculated from the bearing geometry. In the early stages these signals are faint and barely move the overall vibration level, which is why techniques such as envelope analysis, which extract high-frequency impacts, are used.
In practice a single reading rarely gives a definitive diagnosis. Unbalance and misalignment can produce similar 1X peaks; they are told apart by comparing readings in different directions and their phase. So when you plan sensor placement, decide up front which faults you need to distinguish.
To judge whether an overall level is acceptable, industry commonly refers to the ISO 20816 series (formerly ISO 10816). Depending on machine class and mounting, it divides vibration velocity (mm/s RMS) into four zones from A to D: A describes a newly commissioned machine, D a level at which damage is likely. These zones are a starting point; the most useful information is how a machine changes against its own history.
Seven steps to set up vibration monitoring
- Choose the critical machines. Start with assets whose failure stops the line, that have no standby and whose failure history you know: compressors, pumps, fans, gearboxes, main drive motors.
- Define measurement points. Mount the sensor as close to the bearing housing as possible, on a rigid surface. Readings taken through a sheet-metal guard are misleading.
- Match the sensor to the job. Low-cost MEMS accelerometers can be enough for overall level, unbalance and misalignment; early bearing damage needs a wider frequency range and a higher sampling rate. Sample at no less than twice the highest frequency of interest (the Nyquist rule). In dusty, oily or wash-down areas, the sensor's ingress protection is part of the choice too; see our guide to IP65 and IP67 ratings. How high-rate raw data is summarised on site so that only meaningful values travel upstream is covered in edge computing in manufacturing.
- Decide between continuous and periodic. Continuous (online) monitoring suits critical or variable-load machines; periodic route-based readings with a portable collector suit the rest. Connectivity can be wired, Wi-Fi or LPWAN; we compare the options in LoRa vs NB-IoT vs cellular.
- Record a baseline. Collect a few weeks of data while the machine is healthy, across its normal loads and speeds. Alarm limits only mean something against that reference.
- Tier your alarms. Use at least two levels, such as alert and alarm, and write down who receives each and how quickly they must respond. Alongside fixed limits, anomaly detection models that learn a machine's normal behaviour help cut false alarms.
- Link every alarm to a work order. An alarm that does not become a work order in the maintenance management system (CMMS) gets lost. Record what was done and what damage was found so the model improves over time. Keeping service reports and photos of damaged parts in a shared, versioned space such as SmartFiles in Smart360 makes later analysis easier.
Older machines are not an obstacle. External sensors can be fitted to a pump with no controller or a twenty-year-old gearbox; we cover the methods in retrofitting legacy machines.
Common mistakes
- Watching only the overall level. A single mm/s value can catch unbalance but often misses early bearing damage. Without spectrum and envelope analysis, "all normal" is premature.
- Ignoring operating conditions. The same machine vibrates differently at idle and at full load. If speed and load are not stored with each reading, the trend misleads.
- Poor sensor mounting. Magnetic mounts are quick, but their high-frequency response is limited compared with a stud mount; plan the mounting method for permanent installations.
- Leaving the data on a screen nobody watches. When vibration data does not sit alongside OEE and downtime reasons, it is hard to show management its value.
How we do this at Digital Bridge
We treat vibration monitoring as a system built around your machines rather than a box off the shelf:
- We start with a pilot. In our predictive maintenance projects we pilot on one to three critical machines, and the value becomes visible in metrics within the first 30–60 days. Before the pilot we offer a free Industry 4.0 site assessment.
- We design the hardware when needed. If an off-the-shelf sensor does not suit the site, the same team handles electronics and PCB design, mechanical enclosure and embedded firmware. Technical feasibility for device manufacturing and IoT work is free of charge.
- We make the data meaningful. On top of fixed limits we add anomaly detection models that flag departures from each machine's own normal.
- We connect it to what you already run. Vibration data flows into your SCADA and HMI screens or, for multi-site groups, a remote monitoring platform, and alarms land as work orders in the system your maintenance team uses. When we build a bespoke SCADA, you own the source code.
If you want to understand the architecture behind remote monitoring, our IoT remote monitoring platform article is a good place to start. Our other maintenance, quality and energy guides are gathered in our complete Industry 4.0 guide.
Your next step
You do not need a large budget to begin. List the three machines that caused the most downtime in the last 12 months, the cause of each failure and roughly how long each stop lasted. That list largely decides where the pilot should go. Then get in touch: we will walk the site with you, agree the measurement points and set out the pilot scope in a written proposal.