Vibration Alarm Thresholds: How to Set ISO Standards
Key Takeaways
- ISO 10816/20816 defines four severity zones (A, B, C, D) with RMS velocity limits that vary by machine class and mounting type.
- Alert thresholds signal the need for increased monitoring; danger thresholds require a planned or immediate shutdown.
- Statistical thresholds (mean + 2 or 3 standard deviations) adapt to individual machine behavior and significantly reduce nuisance alarms compared to fixed absolute values.
- Band alarms and spectral alarms detect specific fault types earlier than overall RMS alarms alone.
- Thresholds should be reviewed after any significant change in operating conditions, bearing replacements, or major overhauls.
What Are Vibration Alarm Thresholds?
Vibration alarm thresholds define the boundaries between normal operation, elevated risk, and imminent failure for a monitored machine. When a sensor reading crosses a threshold, the monitoring system generates an alert, logs the event, and initiates a predefined response ranging from a maintenance notification to an automatic shutdown command.
Thresholds exist at two distinct levels. The alert (or warning) level is crossed first. It indicates that vibration has risen beyond the normal operating band but that the machine is not yet in danger of immediate failure. The danger level sits higher, typically 1.5 to 2.5 times the alert value, and indicates a condition severe enough to risk catastrophic damage if operation continues.
ISO 10816 and ISO 20816 Severity Zones
The international standard for evaluating machine vibration on non-rotating parts is ISO 10816 (now being superseded by ISO 20816 for new installations). Both standards divide the operational vibration range into four zones based on overall RMS velocity measured in mm/s.
Zone A: Vibration typical of a newly commissioned machine. The machine is in excellent condition.
Zone B: Vibration acceptable for unrestricted long-term operation. No corrective action needed.
Zone C: Vibration unsuitable for continuous long-term operation. The machine can continue running temporarily, but a corrective maintenance plan should be initiated. This zone corresponds broadly to the alert threshold level.
Zone D: Vibration severe enough to cause damage to the machine. Operation is not permitted, and the machine should be taken offline. This zone corresponds to the danger threshold level.
The exact RMS velocity values that define these zones depend on the machine class. The table below shows representative threshold values for the most common classes under ISO 10816-3, which covers industrial machines with power above 15 kW.
| Machine Class | Description | Zone A/B Boundary (mm/s RMS) | Zone B/C Boundary (mm/s RMS) | Zone C/D Boundary (mm/s RMS) |
|---|---|---|---|---|
| Class I | Small machines (up to 15 kW) | 0.71 | 1.8 | 4.5 |
| Class II | Medium machines (15-75 kW) and large machines on rigid mountings | 1.12 | 2.8 | 7.1 |
| Class III | Large machines on rigid mountings (above 75 kW) | 1.8 | 4.5 | 11.2 |
| Class IV | Large machines on flexible mountings (above 75 kW) | 2.8 | 7.1 | 18.0 |
Rigid mountings are those where the machine's natural frequency exceeds its operating speed. Flexible mountings are those where the natural frequency is below operating speed. This distinction matters significantly for threshold selection, since machines on flexible mounts naturally exhibit higher vibration amplitudes at the same fault severity level.
Alert vs. Danger Alarm Levels
Vibration monitoring systems use two distinct alarm levels because not every threshold crossing requires the same response.
The alert level (also called warning level) signals that something has changed. It triggers increased monitoring frequency, inspection scheduling, and investigation into the source. Production does not stop. The machine is watched more closely. This corresponds to the Zone B/C boundary in ISO terminology.
The danger level signals that the machine is at high risk of imminent failure. It triggers a planned or immediate shutdown depending on the criticality of the asset and the available production context. This corresponds to the Zone C/D boundary.
A common industry practice is to set the danger level at 2.5 times the alert level for the same measurement point. This ratio ensures enough separation between the two levels to allow time for investigation and response after the alert fires, before the danger threshold is reached.
Absolute vs. Relative (Baseline) Thresholds
Vibration alarm thresholds fall into two categories: absolute and relative.
Absolute thresholds are fixed values taken directly from ISO standards or OEM specifications. They do not depend on the machine's individual history. The advantage is that they can be applied immediately without collecting baseline data. The disadvantage is that some machines naturally run at vibration levels that differ from the standard's assumptions, leading to either chronic false alarms or under-detection.
Relative (baseline) thresholds are set from the machine's own operating history. They account for the fact that a machine running normally at 3.5 mm/s RMS should have a threshold that reflects its actual normal, not the ISO zone boundary. Relative thresholds require a sufficient baseline period (typically 30 to 90 days of stable operation) but deliver higher sensitivity and fewer nuisance alarms.
In practice, most reliable programs use both: absolute thresholds as a safety floor and relative thresholds as the primary alarm driver. This combination ensures that a machine with an unusually high normal vibration level is still alarmed if it rises further, while also ensuring that it does not cross a danger threshold that was set for a different machine class.
Statistical Alarm Setting: Mean + 2 or 3 Standard Deviations
Statistical alarm setting is the most rigorous approach to relative threshold calibration. The method uses a dataset from the machine's normal operation to establish a statistical envelope of expected behavior.
The process works as follows:
- Collect vibration data from the machine operating under stable, healthy conditions for at least 30 days. Longer baselines (60-90 days) improve accuracy, especially for machines with variable load cycles.
- Calculate the mean (average) and standard deviation of the dataset for each measurement point.
- Set the alert threshold at mean plus 2 standard deviations. This level captures approximately 97.7% of normal variation, meaning only about 2.3% of healthy readings will exceed it.
- Set the danger threshold at mean plus 3 standard deviations. This level captures 99.7% of normal variation, leaving only about 0.3% of false alarms from normal operating variation.
The mean + 2 sigma alert level provides earlier detection of developing faults, while the mean + 3 sigma danger level keeps the false alarm rate low enough to maintain operator confidence in the system. If the false alarm rate at 2 sigma is still too high for a particular asset, teams often shift the alert to 2.5 sigma as a practical compromise.
Statistical thresholds must be recalculated any time operating conditions change significantly. A pump that shifts from 60% load to 100% load will have a new mean and new standard deviation that make the old thresholds invalid.
Types of Vibration Alarms
Vibration analysis systems support multiple alarm types, each suited to detecting different categories of fault.
Overall RMS alarms monitor the total vibration energy across the full measurement frequency range. They are the simplest alarm type and are most effective for detecting gross mechanical problems such as severe unbalance, looseness, or advanced bearing failure. They are less sensitive to early-stage faults because a small fault signal is diluted by the energy from normal machine operation.
Band alarms monitor vibration energy within a defined frequency window. By isolating the frequency range where a specific fault is expected to appear, band alarms detect that fault much earlier than an overall alarm. For example, a band alarm centered on the bearing defect frequency of an outer race fault will respond to that fault well before it contributes enough energy to trigger the overall RMS alarm. Band alarms require knowledge of the machine's fault frequencies, which can be calculated from bearing geometry and shaft speed.
Spectral (narrowband) alarms track the amplitude at specific frequency lines in the vibration spectrum. They are the most precise alarm type and are used to monitor individual harmonics of a known fault frequency. They require the most configuration and are typically used for high-value or high-consequence machines.
Peak and crest factor alarms monitor impulsive energy, which is characteristic of early-stage bearing faults. The crest factor (ratio of peak value to RMS) rises before the overall RMS level does, making these alarms useful for detecting bearing defects before they progress.
Machine Class Definitions
Selecting the correct ISO machine class is a prerequisite for applying any ISO-based threshold. The four classes in ISO 10816-3 are defined by power rating and mounting type.
Class I covers small machines up to 15 kW, including small electric motors and pumps. Class II covers medium-sized machines from 15 to 75 kW, plus larger machines on rigid mountings where the structure stiffness prevents amplification of machine vibration. Class III covers large machines above 75 kW on rigid mountings, including large pumps, fans, and compressors mounted directly on concrete foundations. Class IV covers large machines above 75 kW on flexible mountings, such as machines on spring isolators or on structures with resonant frequencies below operating speed.
Misclassifying a machine is a common source of threshold calibration errors. A Class II threshold applied to a Class IV machine will cause continuous false alarms because the flexible mounting naturally produces higher vibration levels even in a healthy machine.
Worked Example: Setting Alarms for a 15 kW Centrifugal Pump
Consider a 15 kW centrifugal water pump, horizontally mounted on a rigid concrete plinth, running at 1,480 RPM. The pump is classified as ISO Class I (at the boundary of Class I and II based on power rating). After discussion with the site reliability engineer, the team classifies it as Class II given the rigid mounting and similarity of its operating dynamics to larger equipment in the same process line.
Step 1: Determine the ISO zone boundaries for Class II.
- Zone A/B boundary: 1.12 mm/s RMS (new machine acceptance limit)
- Zone B/C boundary: 2.8 mm/s RMS (alert threshold)
- Zone C/D boundary: 7.1 mm/s RMS (danger threshold)
Step 2: Collect 60 days of baseline data from the pump in healthy operation. The mean overall RMS at the drive-end bearing housing is 1.6 mm/s. The standard deviation is 0.3 mm/s.
Step 3: Calculate statistical thresholds.
- Statistical alert: 1.6 + (2 x 0.3) = 2.2 mm/s RMS
- Statistical danger: 1.6 + (3 x 0.3) = 2.5 mm/s RMS
Step 4: Compare statistical thresholds to ISO zone boundaries. The statistical alert (2.2 mm/s) is below the ISO zone B/C boundary (2.8 mm/s), which means the statistical threshold will catch problems earlier than the ISO limit. The statistical danger (2.5 mm/s) is also below the ISO B/C boundary, which is very low separation between alert and danger. The team reviews this and decides to set the danger threshold at the higher of: statistical danger or the ISO B/C boundary. Final settings:
- Alert: 2.2 mm/s RMS (statistical, more sensitive than ISO)
- Danger: 2.8 mm/s RMS (ISO B/C boundary, provides more separation from alert)
This approach combines the sensitivity of statistical thresholds with the safety floor provided by ISO limits, which is the standard practice for condition-based maintenance programs.
The False Alarm Problem and Nuisance Trip Prevention
The most common failure mode of a vibration alarm system is not missed faults: it is excessive false alarms. When a monitoring system generates too many alarms that turn out to be unrelated to real faults, maintenance teams begin to ignore them. This "alarm fatigue" is well-documented in industrial monitoring literature and directly undermines the purpose of the system.
Common sources of false alarms include:
- Thresholds set too tightly relative to normal process variation
- Load changes or speed changes that move the machine's vibration outside a fixed threshold set at a different operating point
- Sensor issues such as loose mounts, cable noise, or electrical interference
- Transient events such as startup, shutdown, or valve operations that produce brief vibration spikes
- Environmental sources such as adjacent machinery, pipes, or structural resonances feeding vibration into the sensor path
Practical nuisance trip prevention strategies include:
- Using time delays: require the threshold to be exceeded for a minimum duration (e.g., 30 seconds) before triggering an alarm
- Using confirmation windows: require the threshold to be exceeded in multiple consecutive readings
- Load normalization: adjust thresholds dynamically as a function of measured process variables such as flow rate, pressure, or motor current
- Excluding startup and shutdown periods from alarm evaluation
- Statistical methods that set thresholds wide enough to exclude normal variation
Anomaly detection algorithms in modern monitoring platforms take this further by learning normal machine behavior across multiple process states and only flagging readings that fall outside the learned pattern for a given operating condition.
Adaptive Alarm Setting
Adaptive alarm setting automates the threshold calibration process by continuously recalculating thresholds from recent operating data. Rather than a static threshold set once and reviewed annually, an adaptive system updates its statistical envelope on a rolling basis, typically using a 30 to 90 day window of recent data.
The practical benefit is that thresholds automatically follow slow changes in machine behavior, such as gradual foundation settling, minor process parameter drift, or seasonal temperature changes that affect bearing clearances. Without adaptation, these slow changes would eventually push normal vibration above a static alert threshold, generating chronic false alarms until someone manually recalibrates.
Adaptive systems include safeguards to prevent the system from learning a deteriorating machine as "normal." The most common safeguard is a rate-of-change limit: if vibration rises faster than a defined rate per day, the adaptive algorithm pauses and flags the machine for human review rather than absorbing the change into the new baseline.
The vibration sensor hardware must be capable of capturing data at the frequency and resolution required for the alarm type. For overall RMS alarms, a basic MEMS accelerometer with periodic data collection is sufficient. For spectral alarms and band alarms, continuous wideband data collection and local FFT processing are required.
Vibration Alarms in a Predictive Maintenance Program
Vibration alarm thresholds are most valuable when they are part of a broader predictive maintenance program rather than a standalone shutdown system. Used in isolation, an alarm only tells you that a problem exists. Integrated with trend data, work order management, and fault diagnostic tools, it tells you what the problem is, how fast it is developing, and when intervention is required.
The standard response workflow for a threshold crossing is: alarm fires, maintenance team investigates using spectrum analysis and other diagnostic data, root cause is identified, a work order is generated with the appropriate priority, and the repair is scheduled before the danger threshold is reached. This workflow keeps machines running longer while preventing unexpected failures.
Asset condition monitoring platforms that combine continuous vibration data with temperature, current, and other signals enable more accurate fault diagnosis from threshold crossings. A vibration alarm accompanied by a rising temperature trend and elevated motor current points toward a bearing problem far more precisely than vibration data alone.
The Bottom Line
Vibration alarm thresholds translate raw vibration data into actionable maintenance decisions. ISO 10816 and ISO 20816 provide the foundational zone boundaries by machine class and mounting type, giving teams a defensible starting point that does not require extensive historical data. Statistical threshold setting using mean plus two or three standard deviations from a machine-specific baseline improves sensitivity and reduces nuisance alarms for assets with enough operating history to support the analysis.
Effective alarm management requires choosing the right alarm type for the fault being detected: overall RMS for gross mechanical problems, band alarms for specific fault frequencies, and spectral alarms for high-value assets where early detection is critical. Separating alert and danger levels with enough margin to allow investigation and response time is as important as setting the correct absolute values.
The practical goal is a system that generates alarms maintenance teams trust and act on, not one they learn to ignore. Getting threshold calibration right, using time delays and confirmation logic to suppress transients, and reviewing thresholds after every significant operational change are the core disciplines that determine whether a vibration monitoring program prevents failures or merely records them.
Set Vibration Alarm Thresholds With Confidence
Tractian's condition monitoring platform applies statistical alarm setting, band alarms, and adaptive thresholds across all your rotating assets, with fault diagnostics built in to explain every alert.
See How Tractian Sets Smart Vibration AlarmsFrequently Asked Questions
What is the difference between an alert alarm and a danger alarm in vibration monitoring?
An alert alarm (sometimes called a warning alarm) signals that vibration has risen above normal operating levels and that the machine needs closer attention. It does not require an immediate shutdown. A danger alarm represents a significantly higher threshold, typically 1.5 to 2.5 times the alert level, that indicates imminent risk of failure. When a danger alarm fires, the standard response is to take the machine offline as soon as production allows.
What are the ISO 10816 vibration severity zones?
ISO 10816 defines four severity zones. Zone A covers newly commissioned machines in good condition. Zone B is acceptable for long-term operation. Zone C is tolerable for a limited time only, and corrective action should be planned. Zone D represents vibration severe enough to cause machine damage, and operation in this zone is not permitted. Each zone is defined by an RMS velocity range in mm/s, with exact values depending on machine class.
What is the difference between absolute and relative vibration alarm thresholds?
Absolute thresholds are fixed values derived from ISO standards or OEM specifications and apply regardless of the machine's individual history. Relative thresholds are set from the machine's own baseline, typically using a statistical method such as mean plus two or three standard deviations from normal operating data. Absolute thresholds are easier to apply without historical data, while relative thresholds are more sensitive to changes specific to that machine and produce fewer false alarms on assets that run at non-standard vibration levels.
How do you set vibration alarm thresholds using the statistical method?
Collect at least 30 days of vibration data from the machine running in its normal, healthy state. Calculate the mean and standard deviation of that dataset. Set the alert threshold at mean plus 2 standard deviations (captures roughly 97.7% of normal variation) and the danger threshold at mean plus 3 standard deviations (captures 99.7%). Review and recalibrate after any significant change to operating conditions such as load, speed, or process fluid.
What is a band alarm in vibration monitoring?
A band alarm monitors vibration energy within a defined frequency range rather than the full-spectrum overall RMS level. By isolating a narrow frequency band, teams can track specific fault signatures such as bearing defect frequencies, gear mesh frequencies, or unbalance without broad vibration masking the signal. Band alarms enable earlier detection of specific faults than overall RMS alarms alone.
Why do vibration alarm thresholds differ by machine class?
Machine class accounts for the fact that a large, rigidly mounted 500 kW motor naturally produces different vibration levels than a small pump mounted on a flexible frame. ISO 10816 and ISO 20816 define machine classes based on power rating and mounting type, with correspondingly different zone boundaries. Applying a threshold designed for a large rigid machine to a small flexible one would result in either constant false alarms or missed real faults.
How often should vibration alarm thresholds be reviewed?
Thresholds should be reviewed after any significant change in operating conditions: load changes, speed changes, process fluid changes, bearing replacements, or major overhauls. For machines with stable operating conditions, an annual review is a reasonable minimum. Adaptive alarm systems automate this by continuously recalculating thresholds from recent data.
Related terms
Equipment Health Index: Definition
An equipment health index aggregates sensor data into a single condition score. Learn how EHI is calculated, how it differs from OEE, and how it supports predictive maintenance decisions.
Equipment Monitoring: Definition
Equipment monitoring measures key machine parameters in real time to detect faults before failure. Learn how it works, the types of monitoring, and how it differs from condition monitoring.
Forward Scheduling: Definition
Forward scheduling starts from today and assigns tasks forward in time. Learn how it works, how it compares to backward scheduling, and when to use it in maintenance and production planning.
Expected Useful Life: Definition
Expected useful life estimates how long an asset will remain functional under normal conditions. Learn how it is determined, how it affects depreciation, and its role in replace-or-repair decisions.
Facility Manager: Role
A facility manager oversees buildings, maintenance, compliance, and space to support operations. Learn responsibilities, CFM certification, tools like CMMS and CAFM, and key KPIs.