For most long-term infrastructure projects, the best starting point is a hybrid architecture: the system observes the condition continuously, stores lightweight periodic results, and after a trigger keeps the full waveform with a pre-event segment. Continuous recording of the full waveform makes sense when you cannot define the event in advance, you need modal analysis, or losing a short impulse would have an unacceptable consequence. Choose event-driven mode only after testing trigger sensitivity, buffer size, local memory, and behavior during a transmission outage.
In brief
- “Continuous monitoring” can mean always-on observation, periodic results, or uninterrupted raw waveform recording. These variants have very different costs.
- A trigger reduces data volume, but it introduces the risk of missing an event, clipping the start, and overwhelming the system with events.
- TCO includes not only devices and licenses. Count installation, access, transmission, storage, calibration, event review, service, data loss, and system exit.
- Acceptance should cover the full chain: excitation, recording, completeness, analysis, alarm, notification, and proof of response.
- Inclify handles numerical data and full dynamic events in parallel. This does not change the limits of the sensor, trigger, or field connection.
In offers, the word “continuous” is exceptionally flexible. One supplier means a sensor that waits around the clock for a threshold crossing. Another sends a maximum value at a fixed interval, for example every 15 minutes. A third stores every sample on all axes. Each can use the same heading, even though you are buying three different capabilities.
That difference only becomes visible in the first dispute. The engineer asks for ten seconds of waveform before the impact, and the system has only a maximum value. The manager wants to check weaker events that preceded the threshold crossing, but the trigger did not record them. IT, in turn, discovers that “full history” means hundreds of gigabytes with no agreed retention or export rule.
So do not start by choosing the device. First write down which decision cannot be late and which evidence you will need after the event.
Four modes that the market puts into one basket
Continuous observation means that the measurement chain stays active and can detect a change at any time. It does not decide whether all samples are stored.
Continuous full-waveform recording stores consecutive waveform samples without gaps within a defined band and resolution. Such material lets you return to any moment, but it requires memory, transmission, completeness control, and retention rules.
Periodic recording stores a result at intervals, for example maximum, RMS, or selected bands from the last minute. It shows the trend and operating condition well. It does not reconstruct the shape of an impulse between points if the device did not keep the waveform locally.
Event-driven recording creates a record after a trigger condition is met. A good record includes pre-trigger, the trigger moment, post-trigger, all required axes, and metadata. A weak record starts after the most important peak.
Hybrid architecture combines a lightweight periodic layer with full event records. In the background, you see cadence, chain health, and level changes, and after a meaningful excitation you get a waveform for analysis. It is not automatically the best solution, but it usually balances evidential material and cost best.
| Mode | What you keep | Strongest point | Main risk | Typical reason to buy |
|---|---|---|---|---|
| continuous full waveform | every sample over the entire period | analysis of phenomena that could not be predicted | volume, transfer, retention, and review | research, diagnostics, critical asset |
| periodic | aggregates at a fixed cadence | low-cost trend and availability supervision | loss of short peaks and event shape | operation with slow changes |
| event-driven | records triggered by a trigger | full material for selected events | missing or clipping the event | impulse works, blasting, traffic passes |
| hybrid | periodic trend and full events | context before, between, and after events | more demanding configuration and testing | long-term construction and infrastructure monitoring |
NI distinguishes online monitoring, where data are collected continuously, from periodic site visits with a portable sensor. That is a useful operational split, but in the specification you still need to state which part of the raw signal goes into storage. Device manufacturers also show several operating modes side by side: alarm, periodic, and trigger-based. Having a mode does not mean it fits your standard or response time.
First define the loss you cannot accept
ISO 4866 treats frequency, duration, and response amplitude as parameters important for assessing structural vibration. The excitation source also affects the required dynamic and frequency range. In purchasing terms: you cannot honestly choose an architecture without describing the phenomenon.
Ask five questions:
- Can the event be repeated under control, or does it occur once and disappear?
- Does the decision depend on PPV, RMS, FFT, third-octave bands, duration, impulse sequence, or the response of several points?
- How early must the system react: seconds, minutes, the next work shift?
- Does a missing record mean only a lack of knowledge, or can it stop the works or weaken your position in a dispute?
- Who will review the material, and how many events per day can that person assess reliably?
If a one-off event can determine a claim, design for the loss of the worst record, not for the average day. When the phenomenon is slow and well represented by periodic readings, a full waveform over 24 hours may not add any decision value. For works that change character every week, a hybrid setup lets you tune the trigger against the background without giving up the trend.
When full continuous recording justifies its cost
Continuous full-waveform recording is sensible when the trigger does not have a stable definition. This applies to studies of new sources, identification of natural frequencies, analysis of the response to multiple overlapping excitations, or a startup period in which you are still learning the background. It may also be required when the evidential procedure demands uninterrupted material.
There is also a less obvious case: the event is weak at the trigger point, but important elsewhere. A local trigger on one axis may not open the records needed to compare phase and propagation. In that case, central triggering, synchronized devices, or continuous recording during a high-risk period may be safer.
Do not assume, however, that continuous recording removes every gap. The device may overfill memory, the link may fail to transfer data, clocks may drift apart, and the retention process may delete material before analysis. A full waveform without agreed completeness metrics and data SLA is only a large set of unknown quality.
When event-driven mode is really enough
Event-driven mode works well when the important excitation has a repeatable input signal, the measurement chain has the right range, and the threshold can be set below the level that requires reaction. You also need a pre-trigger. Its role is described in detail in the guide on sampling frequency, trigger, and pre-trigger.
The “below alarm” condition matters. The trigger is for recording, the alarm is for decision-making. If both thresholds are identical, you will not collect weaker events for model calibration or evidence of escalation. If the trigger is too low, every strike on the enclosure can start a record, a transmission, and a review process.
Before acceptance, check at least:
- the filter and the quantity used by the trigger;
- the entry threshold, hold time, and re-arm condition;
- the buffer length before and after triggering;
- simultaneous recording of all required axes;
- behavior when two events occur close together;
- local queue capacity without connectivity;
- recognition of duplicates and missing fragments;
- how an incomplete record is marked.
Without these parameters, “event recording” is a label, not an acceptance requirement.
TCO: count people and invisible risk too
Total vibration monitoring cost can be written as:
TCO = equipment + installation + access + transmission + platform + storage + calibration + service + analysis + alarm handling + downtime + data migration and release
Not every component grows in the same way. Full recording increases storage and analysis cost, but it may reduce disputes about missing evidence. A very sensitive trigger lowers the risk of missing an event, but it increases the number of records to review. A low-cost battery sensor may reduce cable cost, while at the same time requiring more frequent power-source replacement under intensive operation.
| TCO component | What drives it | Question for the offer |
|---|---|---|
| installation and access | number of points, height, drilling, traffic closures | how much it costs to re-enter the point |
| transmission | number of samples, events, protocol overhead, retransmissions | what happens without coverage |
| storage | full waveforms, retention period, copies, and indexes | what exactly is subject to retention |
| analysis | number of records and metric complexity | who reviews the event and under which SLA |
| maintenance | battery, calibration, sealing, cabling | what the visit and spare-part plan is |
| alarms | false triggers, number of recipients, out-of-hours work | how many test alerts go through the procedure |
| data loss | missing record, incompleteness, wrong time | how the system exposes the gap before a report is created |
| exit | export format, documentation, post-contract access | how much a full history release costs |
A broader expense model is available in the article on TCO for structural monitoring. In a vibration project, add the cost of reviewing full waveforms. One hundred automatic events per day is not cheap if each one requires fifteen minutes of specialist time.
Numerical example: waveform versus events
Assume six three-axis points, 256 Hz sampling, and 2 bytes per sample. We do not add headers, metadata, indexes, or copies. This is intentionally a simple raw-volume calculation.
A full waveform over one day gives:
6 points × 3 axes × 256 samples/s × 86 400 s × 2 B = 796 262 400 B
That is about 759 MiB per day. Over 30 days, the raw material will take about 22.2 GiB before system overhead is added.
If the hybrid setup stores 120 records per day, each 30 seconds long, the event part is:
6 × 3 × 256 × 30 × 2 × 120 = 33 177 600 B
That is about 31.6 MiB per day. A light periodic trend is added, but the difference in volume is still large. This is not an argument for cutting data at any cost. It shows the price of retention decisions and requires a test of whether 120 records cover all needed events.
Now count the work. If an operator reviews each of the 120 records for three minutes, that consumes six hours per day. Automatic classification and ranking help, but the acceptance criterion still has to show which records need a human. The most expensive variant can therefore come from cheap transmission and a badly set trigger.
Decision matrix by scenario
| Scenario | Reasonable starting point | Why | What to verify |
|---|---|---|---|
| piling next to a neighboring building | hybrid | work trend and full impulses for analysis | trigger, pre-trigger, PPV or the required metric |
| investigation of an unknown vibration source | full recording in a diagnostic campaign | you do not yet know the event signature | retention, synchronization, and band |
| long-term supervision of a machine with stable operation | periodic plus events | RMS trend and waveforms after change | operating modes, load, and alarms |
| load test or modal test | full waveform during the test window | phase, time, and response of multiple points are needed | common time, calibration, excitation |
| remote battery point with poor connectivity | event-driven or hybrid with local memory | energy and transmission limits | buffer, queue, retry, battery |
| asset with slow changes and no impulses | periodic | full waveform does not add a decision | cadence versus phenomenon dynamics |
This is a starting point, not an automatic recipe. The same site may move through several modes. During the exploratory week, you record more. After the background is established, you switch to hybrid. During a particularly risky operation, you enable extended recording again. Require the configuration change to be planned and documented.
How to write requirements so offers are comparable
Instead of “the system must work continuously,” write a measurable contract:
- band, sampling frequency, and input unit;
- number of axes, orientation, and time synchronization;
- amplitude range, noise level, and anti-aliasing filter;
- definition of the periodic aggregate together with the window;
- trigger, actuation tolerance, pre-trigger, and post-trigger;
- maximum number of events without loss and local buffering time;
- identification of fragments, duplicates, and incomplete records;
- required time from event to alarm and data visibility;
- export format for full waveforms, results, and metadata;
- retention period separately for waveform, aggregates, and communication logs;
- procedure for changing parameters and the responsible person;
- FAT and SAT test set with pass criteria.
Such a description exposes the difference between a system that measures continuously and a system that continuously delivers material for decisions.
30-day pilot checklist
- [ ] Background was recorded before the main works started.
- [ ] Controlled events were executed below, at, and above the trigger.
- [ ] Every record has the expected length and pre-trigger.
- [ ] Completeness of all axes and fragments was verified.
- [ ] A communication outage does not delete records from the device.
- [ ] Resending does not create a second event.
- [ ] The periodic trend and the full waveform share the same UTC time.
- [ ] The alarm reaches the right people within the recorded time.
- [ ] The number of false and missed events was measured.
- [ ] Daily volume and specialist review time were measured.
- [ ] The export contains the waveform, parameters, and completeness status.
- [ ] Retention and the procedure for securing a disputed event were defined.
How Inclify supports this
Inclify accepts ordinary numerical readings and dynamic events as separate data families. That allows one project to show a lightweight operating trend and full vibration records sent after triggering. The system does not require both streams to have the same cadence.
For a dynamic record, the platform tracks the expected and received number of fragments. It shows progress, and a record that can no longer be completed may end with a clear incomplete status. Silence in transmission alone does not close the event, because an overloaded device may resume sending later. This matters in an event-driven architecture, where the absence of a chart must not pretend to be the absence of vibration.
After a successful analysis, the available outputs are the time waveform, FFT, and 21 third-octave bands from 1 to 100 Hz. The profile with 4th-order Butterworth filters determines RMS and MAX in the bands, requires an effective sampling rate above 224 Hz, and at least 10 s of pre-trigger. The dynamic alarm compares active bands with the reference envelope and uses warning and alarm levels with hysteresis. Notifications can be sent by email and SMS, according to user preferences.
The administrator also has access to compressed logs of exchanges with devices for the configured period, by default seven days. The log helps separate a sensor problem, a mapping problem, and a transmission problem. It is not a multi-year archive of the full waveform.
Inclify does not set the trigger inside any arbitrary recorder and does not increase its local memory. It cannot recover an event that the device did not keep. The platform also does not claim automatic compliance of the entire chain with a selected standard. The equipment, installation, filter, timing, and procedure are accepted separately.
Limitations and red flags
The first red flag is an offer that gives only a sending interval. “Data every 30 seconds” does not say whether the content is a maximum, RMS, waveform fragment, or last sample. The second is a missing pre-trigger definition. The third is a chart without completeness status.
Also watch out for retention described with a single period. A company may keep aggregates for five years, but waveforms only for one week. Or the other way around, keep the files but delete the configuration needed to recalculate them. Write raw waveform, results, metadata, communication logs, and decision audit separately.
A hybrid setup is not maintenance-free. Someone has to review the trigger distribution, background changes, battery, and gaps. If no process owner exists, the system will over time produce too many events or lose sensitivity after operating conditions change.
FAQ
Can event-driven monitoring be called continuous?
You can say the device is always on, but you must add that it records the full waveform only after a trigger. In the contract, avoid the word “continuous” on its own. Define what is measured without interruption, what is stored periodically, and what is kept only after triggering. Otherwise the parties may expect different material.
Does full continuous recording always provide the best evidence?
No. It must be complete, calibrated, synchronized, and preserved together with the configuration. A huge file without axis information, filter, mounting, and missing samples may be weaker than a well-described event record. Full recording lowers the risk of missing the moment, but it does not repair chain quality.
How do I set the trigger so I do not miss an alarm?
The setting comes from the background, noise, the quantity used for detection, and the level that requires reaction. The trigger threshold should usually leave some margin below the alarm threshold, but there is no single percentage for all assets. Run tests below, at, and above the threshold, then compare detected events with an independent reference record.
Which is more expensive: continuous recording or a hybrid setup?
For storage and transmission, full recording is usually more expensive. A hybrid setup can still lose on cost if a badly set trigger creates thousands of records and requires manual review. Compare TCO on a real pilot: volume, visits, batteries, specialist time, false alarms, and the cost of data loss.
Is a hybrid architecture suitable for compliance assessment?
It can be, if it preserves the quantity, band, time, point, completeness, and processing method required by the relevant document. Trend plus record alone does not mean compliance. A specialist must confirm the sensor, mounting, sampling, filtering, metric, and thresholds for the specific purpose.
How long should full vibration waveforms be kept?
The period should follow the contract, analysis time, possible claims, project requirements, and cost. A record linked to an alarm or a dispute usually needs separate protection from routine retention. Do not assume that the period for trends, transmission logs, and raw waveforms is the same. Each one must be written into the data policy.
Sources and further reading
- ISO 4866:2010, Vibration of fixed structures - principles of measurement and data processing, taking frequency, duration, and amplitude into account.
- NI, What is Vibration Monitoring? - distinction between online monitoring and periodic field measurements, plus an overview of analysis methods.
- Worldsensing, Vibration Meter - official example of a device with different modes, alerts, and waveform storage for selected events.
- IEC 61260-1:2014, octave-band and fractional-octave-band filters - requirements for bandpass filters, important in third-octave analysis.
- NIST Technical Note 334, sampled data and switching logic - fundamentals of sampling and aliasing.
- US Bureau of Reclamation, Design Standards No. 13, Chapter 11 - planning, pre-installation tests, documentation, and maintenance of instrumentation systems.
Next decision
Take seven days of data from the most intense stage and count four numbers: the number of significant events, the number of false triggers, the volume of full records, and the time needed to review them. Without that, TCO is guesswork. Talk to the Inclify team if you want to compare continuous, event-driven, and hybrid options on one real scenario, without replacing the entire installation at the start.