Select vibration sampling frequency, trigger and pre-trigger backwards from the highest frequency, the shortest event, the required metric and the time before the start of the phenomenon. The statement “sample twice as fast as the band” describes a theoretical limit, not a complete design. You still need an anti-aliasing filter, margin, the correct amplitude range and a test of the full signal chain.
In brief
- Nyquist tells you when an ideally band-limited signal can be reconstructed; it does not guarantee correct amplitude at the band edge or the quality of a real recorder.
- Trigger decides which events you save. If set too high, it misses important impulses. If set too low, it creates a flood of records and can overload transmission.
- Pre-trigger preserves data before the trigger. Without it, you will not see the background, the rise or the true start of the impulse.
- For the Inclify third-octave profile with Butterworth filters, effective sampling above 224 Hz and at least 10 s of pre-trigger are required.
- Accept not a menu setting, but the full test waveform, metadata and the analysis result in the relevant band.
The most expensive error in vibration recording appears after the event. The file exists, the timestamp is correct and the chart looks professional. Only then does a specialist notice that the highest band sits too close to the sampling limit, the first half of the impulse has disappeared or the signal has been clipped by the input range. That event cannot be repeated.
That is why the specification “recorder minimum 200 Hz” is too short. It does not say what is to be measured, which filter acts before the converter, how much data the buffer retains, how recording is triggered or what happens when transmission is delayed. A good design describes the path from the phenomenon to the result and then tests it with a signal of known characteristics.
If the result is to support impact assessment for construction works, the starting point is the objective and method described in the guide to building vibration monitoring during construction, not the frequency in the datasheet.
Three parameters, three different tasks
Sampling frequency is the number of evenly taken signal samples per second. If a recorder stores 256 samples per second, the nominal interval is about 3.906 ms. The number alone does not yet tell you the sensor bandwidth, filtering or clock accuracy.
Trigger is the rule that starts event recording when a condition is met, for example an amplitude threshold in a selected axis. It can operate in the recorder before the data reach the platform. It should not be confused with a later alarm, which evaluates already processed results.
Pre-trigger is the part of the waveform before the trigger moment, preserved by a circular buffer. It shows the background, the rise and the start of the event. Post-trigger covers the part after the trigger. Together they form the window needed for analysis.
| Parameter | Protects against | Poor selection causes | Acceptance test |
|---|---|---|---|
| sampling | loss of time and frequency information | aliasing, amplitude error, lack of bandwidth | sine waves and multi-tone signal |
| anti-aliasing filter | folding of frequencies outside the band | false components in the spectrum | tone above the usable band |
| trigger | missing a significant event | no record or too many events | series of impulses above and below threshold |
| pre-trigger | loss of the start and the background | clipped impulse, no reference point | trigger after a known delay |
| input range | saturation and poor resolution | flattened peaks or high noise | amplitudes from background to maximum |
| clock | incorrect source-response alignment | shifted correlations | comparison with a time reference |
Nyquist without the shortcut “twice is enough”
For an ideal signal band-limited to frequency fmax, the sampling condition requires fs > 2 · fmax. Half the sampling frequency is the Nyquist frequency. Components above that limit can fold into a lower band as aliases. NIST shows this mechanism in material on sampled systems: after folding, the false component cannot be distinguished from the real one using samples alone.
In practice, the signal does not end abruptly at fmax, and a filter does not have a vertical characteristic. It needs a transition band between the useful frequency and the area of effective attenuation. The smaller the margin between fmax and fs/2, the sharper the filter must be and the more important its phase, attenuation and tolerances become.
Example: you want to analyse response up to 100 Hz. fs = 200 Hz places the highest relevant frequency exactly on the Nyquist boundary. It leaves no room for the filter transition, and sinusoidal samples may fall in an unfavourable phase. fs = 201 Hz satisfies the simple inequality, but it still does not automatically make the design good. The safety factor comes from the signal chain, the filter and the required accuracy, not from a universal slogan.
IEC 61260-1 defines requirements for analog, sampled and digital band-pass filters and their classes. Compliance of the band-pass algorithm does not follow from the number of samples alone. You must assess the sensor, conditioning, converter, filter, environmental conditions and uncertainty.
Select recording from the final result backwards
The first step is one decision sentence: “I want to assess construction events in bands up to 100 Hz and preserve the start of the impulse” or “I want to identify the dominant frequency of a machine up to 40 Hz.” Next you select the assessment document and the required quantity, acceleration, velocity or displacement. Only then does the list of signal-chain parameters emerge.
The procedure is as follows:
- determine the highest and lowest frequency relevant to the decision;
- define the allowable amplitude and phase error in the band;
- select the sensor and conditioning with the right characteristic;
- design the anti-aliasing filter and the sampling frequency with margin;
- estimate the maximum amplitude and the background to choose the dynamic range;
- define the pre-trigger time, post-trigger time and maximum event duration;
- set the trigger rule and the re-arm condition;
- calculate data volume, throughput and behaviour during a communication break;
- prepare acceptance signals and correctness criteria.
ISO 4866 emphasises that the source affects the required dynamic range, frequency and other parameters. An impact from construction works, continuous machine vibration and vehicle passage should not automatically use the same window. Assessment of damage and assessment of human comfort also leads to different measurement questions. The choice of metric is described in a separate guide on PPV, RMS, FFT and third-octave bands. If the result is velocity derived from an accelerometer, separate requirements apply to integration, filters and drift control.
Trigger: the recording threshold is not a safety threshold
The trigger is meant to provide material for analysis. The safety threshold is used for decisions after the correct metric has been calculated. If you set triggering at the alarm level, you may fail to preserve the events needed to understand the background, tuning and rise. If you set it very low without limits, you will record enclosure taps, maintenance movement and continuous noise.
The trigger rule design should specify:
- the quantity and axis used for detection;
- the detection filter, if any;
- the entry threshold and the end condition;
- hold time and re-arm lockout;
- behaviour when several axes trigger at the same time;
- maximum record duration;
- how test, maintenance and real event are labelled.
A step test is worth running: apply impulses below, near and above the threshold and vary their duration. The criterion should not be “it worked sometimes.” It should define which impulses are to be recorded, what the trigger time tolerance is and whether the pre-trigger preserves the expected segment.
Pre-trigger: the part of the event that has not happened yet
A recorder with pre-trigger continuously stores the newest samples in a buffer. After the trigger, it freezes part of them and appends post-event data. If the trigger or packet transmission is delayed, the buffer must still protect the correct range. Silence in the link does not prove that the event has ended.
Pre-trigger helps answer four questions: what was the background, did the signal rise, when did the impulse begin and did the trigger react to the late part of the phenomenon. It is also needed by filters that have an initial state and a settling time. A segment that is too short can contaminate the beginning of the result with filter transients.
Illustrative example. You want to use the Inclify profile with 21 third-octave bands from 1-100 Hz and Butterworth filters. The profile requires effective sampling above 224 Hz and at least 10 s of pre-trigger. Assume fs = 256 Hz, 10 s before and 20 s after the trigger. You get 2 560 pre-trigger samples and 7 680 samples in a 30-second record on one axis. This is an example of data volume, not a universal setting. The design must still confirm the anti-aliasing filter, amplitude, event duration and standard requirement.
If only 27 s arrive out of 30 s, the record must not be silently called complete. The missing segment may contain the peak. The result should retain the status of incompleteness, the number of expected and received parts, and information on whether the material is suitable for the chosen analysis at all.
Throughput and memory are part of the measurement
Calculate record volume before purchase. Multiply sampling frequency by time, number of axes, sample size and number of events. Add metadata and transmission overhead. This calculation is not only for the IT department. If the link cannot keep up, the device needs a local buffer, a queue and safe retries without duplicating segments.
For event series, priority rules matter. The recorder must not overwrite a record that has not yet been transmitted just because another trigger occurred. The platform should distinguish pending, incomplete, completed and faulty events. The engineer must know whether the missing chart means physical silence, delayed transmission or data loss.
Run the load test as a series of events, not a single impulse. Acceptance should show how many records the device can safely buffer, whether it preserves order and timestamps after the link returns and whether resending the same segment is idempotent. Only then does the sampling parameter become a functioning system capability, not a number in the datasheet. A full comparison of the impact of continuous, event-based and hybrid recording on risk and TCO helps decide what capacity you really need.
It is worth keeping the full record, metadata and test result together, especially when the measurement may later become evidence in a dispute.
Procurement checklist for vibration recording
- [ ] The objective, assessment document and highest band are documented before equipment selection.
- [ ] The supplier provides the sensor and full signal-chain characteristic, not only
fs. - [ ] The anti-aliasing filter, its frequency and attenuation are known.
- [ ] Sampling frequency has justified margin relative to the band.
- [ ] The input range covers the background and the largest expected impulse.
- [ ] Resolution and noise have been assessed in the final unit.
- [ ] The trigger has a definition of threshold, filter, hold and re-arm.
- [ ] Pre-trigger and post-trigger follow from the event duration and filtering.
- [ ] The record preserves exact time, axis orientation and calibration profile.
- [ ] A transmission break does not delete the locally recorded event.
- [ ] An incomplete record has a clear status and does not pretend to be a complete result.
- [ ] Acceptance includes sine waves, impulse, overrange and out-of-band signal.
- [ ] The full waveform can be retained for re-analysis.
- [ ] The procedure distinguishes recorder trigger from safety alarm.
How it works in Inclify
Inclify calculates the effective sampling frequency of the full event from the number of samples and its actual time window. It does not base the evaluation only on the nominal value entered in the configuration. For events it shows the time waveform, FFT and 21 third-octave bands from 1 Hz to 100 Hz.
The profile with 4th-order Butterworth filters requires effective sampling above 224 Hz and at least 10 s of pre-trigger. It calculates RMS and MAX in the bands. Calibration is set separately for each axis, so there is no need to assume one coefficient for the entire sensor.
With split transmission, the platform tracks the number of expected and received segments and the progress. A record may end with incomplete status if later data confirm that older segments will not arrive. A lack of transmission alone does not close the event, because an overloaded device may resume sending. The platform will not fix aliasing, saturation or the lack of pre-trigger that were not recorded in the field.
After correct analysis, an envelope alarm can compare RMS and MAX in active bands with references and apply WARNING / ALARM levels with hysteresis.
Limitations: what you cannot recover after the event
You cannot filter aliasing after the fact in a way that restores the true frequency. You cannot recover a peak clipped by an overdriven converter or the start of an impulse that was outside the buffer. An algorithm may smooth the chart, but it does not create missing evidence.
Nominal 256 Hz does not automatically mean 256 equal intervals. Jitter, a wrong clock, missing samples and a fixed record duration can change the effective frequency. That is why acceptance should compare the number of samples, timestamps and event length, not only the configuration field.
The Inclify profile requirement is not a declaration that the whole chain complies with the standard. IEC 61260-1 covers filter characteristics and classes; ISO 4866 covers broader measurement principles. The correct document may require a different band, point, quantity or procedure. A specialist who knows the objective and the source approves the design.
FAQ
Is sampling frequency twice the band enough?
Not as a general design rule. It is a boundary derived from the sampling theorem for an ideally band-limited signal. A real filter needs a transition band, the sensor has its own characteristic, and the analysis requires a defined amplitude accuracy. Select margin based on the full signal chain and confirm it with a test signal above the band.
How is a trigger different from a vibration alarm?
A trigger starts recording in the recorder and decides whether a full record is created. An alarm works later on values or bands calculated from the recorded event and starts an operational procedure. They may use different quantities and thresholds. Setting the trigger equal to the alarm may deprive you of background data and weaker events.
How many seconds of pre-trigger should be set?
As much as needed to capture the background, the rise, detection delay and filter settling. For the Inclify third-octave profile with Butterworth filters, the minimum is 10 s. Another analysis may require a different window. Base the decision on event duration and the signal chain, not on the smallest memory-saving value.
Does higher sampling always give a better result?
No. It increases volume, transmission load and processing cost, and it will not fix a poor sensor, overrange or bad mounting. It may be needed for the band and filter, but it should be justified. Excess data that prevents timely transmission is also an operational risk. Always accept the whole chain, not one parameter.
How do you identify an incomplete waveform?
Compare the expected time, the number of samples and the received segments for each axis. Check the start, the end and the continuity of the timestamps. The absence of a flat segment in the chart is not enough, because the visualisation may connect points across a gap. The completeness status should be stored next to the result and block unconditional interpretation.
Which signals should be included in the acceptance test?
Use sine waves at several points in the band, a tone near the upper limit, a signal above the band, a short impulse, an amplitude close to saturation and a signal near the noise floor. Add a series of events around the trigger threshold. For each one, record the expected time, amplitude, spectrum, completeness, pre-trigger content and acceptance criterion. The entire path from test rig to object is structured by the FAT and SAT plan for a vibration monitoring system.
Sources and further reading
- NIST Technical Note 334 - An Introduction to Sampled Data and Switching Logic - sampling theorem, reconstruction and aliasing.
- NIST Time and Frequency Division - Section 10: sampled data - aliasing, spectral leakage and practical sampling limits.
- IEC 61260-1:2014 - Octave-band and fractional-octave-band filters - requirements for analog, sampled and digital band-pass filters.
- ISO 4866:2010 - Vibration of fixed structures - selecting measurement by frequency, duration and response amplitude.
- FHWA NHI-09-010 - Road Tunnel Manual, chapter 15 - official guidance on instrumentation and vibration monitoring for tunnels.
What next
Take one current recorder profile and add next to it: decision band, anti-aliasing filter, amplitude range, trigger condition, pre-trigger, post-trigger and acceptance test. Every empty field means a risk of losing a unique event. Talk to the Inclify team if you want to walk through a sample file and verify whether its parameters are sufficient for the planned analysis.