Vibration Analysis Fundamentals for Predictive Maintenance

Vibration Analysis Fundamentals for Predictive Maintenance

Introduction to vibration analysis covering sensor selection, measurement parameters, frequency analysis, fault detection, and ISO 10816 standards. This comprehensive guide covers the essential concepts, practical implementation strategies, and industry best practices that every automation engineer should know. The material below synthesizes practical field experience with published guidance on methodologies, sensor technology, and integration practices to build reliable predictive maintenance (PdM) programs.

Key Concepts

Understanding the fundamentals is critical for successful implementation. Vibration analysis measures mechanical oscillations in rotating and reciprocating equipment (motors, pumps, fans, gearboxes, compressors) to detect faults such as imbalance, misalignment, bearing defects, gear mesh problems, and structural looseness. Equipment faults produce characteristic changes in vibration amplitude and frequency content that trained analysts or automated classifiers can map to root causes.

Analysis Domains and Techniques

Practitioners use three primary analysis domains:

• Time domain — raw vibration vs. time for identifying transients, impacts, start/stop events and overall waveform characteristics. Time-domain analysis is essential for capturing event timing, shock impulses and long-period trends (see Viking Analytics 7-step workflow) [1].
• Frequency domain (FFT) — converts time signals into spectra to reveal energy at discrete frequencies. FFTs expose fault signatures such as 1× rotational (imbalance), 2× (misalignment) and gear-mesh frequencies, and allow RPM-referenced fault libraries to be applied [1][4].
• Envelope (demodulation) analysis — isolates high-frequency resonances excited by repetitive impacts (commonly used for early bearing defect detection). Envelope analysis extracts modulated bearing defect frequencies otherwise masked in the broadband spectrum [1][3][5].

Measurement Parameters and Units

Choose measurement quantity by application:

• Acceleration (m/s² or g): best for high-frequency phenomena and bearing fault detection using envelope analysis; accelerometers are the most common sensor choice.
• Velocity (mm/s RMS): preferred for overall machine condition and severity evaluation per ISO 10816 guidelines; velocity correlates well with perceived vibration severity for typical rotating machinery speeds [1][5].
• Displacement (µm peak-to-peak): useful at very low frequencies and for shaft relative motion measurements (e.g., rubs or looseness).

Sampling frequency must exceed the highest frequency of interest by the Nyquist margin. For bearing envelope and resonance work, bandwidths on the order of several kHz to tens of kHz are common; for overall machine monitoring, lower sampling rates (e.g., a few kHz) may suffice [4][5].

Fault Signatures and Frequency Mapping

Common fault-to-frequency mappings used in diagnostic libraries:

• Imbalance: prominent peak at 1× shaft speed (1× RPM).
• Misalignment: energy at 1× and 2× RPM, often with a phase shift and axial component.
• Looseness / Structural Resonance: broad-band increase, harmonics and sidebands around rotating frequencies.
• Bearing defects: distinct bearing characteristic frequencies (BPFO, BPFI, BSF, FTF) and their harmonics and sidebands; best detected with envelope analysis of high-frequency acceleration [1][3][8].
• Gear problems: gear mesh frequency (GMF) and sidebands; wear and broken teeth manifest as changes in GMF amplitude and modulation patterns.

Accurate mapping requires RPM-referenced analysis and knowledge of machine geometry (bearing element counts, gear teeth, belt sheave ratios). Establishing a baseline spectral library for each asset simplifies automated fault classification [1][3].

Implementation Guide

Successful implementation requires careful planning, proper tool selection, and adherence to industry standards. The following step-by-step process reflects best-practice workflows observed in field deployments and vendor guidance.

1. Asset Prioritization and Initial Assessment

Start by ranking assets by criticality: failure cost, downtime impact, past reliability, and failure modes. Prioritize high-failure-cost machines such as centrifugal pumps, induction motors, fans, compressors and gearboxes. Document nameplate data, typical operating RPM bands, couplings, bearing types and service loads. This data supports correct sensor placement, sampling settings, and fault frequency calculations [2][3].

2. Sensor Selection and Placement

Select accelerometers for general PdM due to sensitivity across frequencies; choose sensors rated for the expected amplitude range and environmental conditions (temperature, ingress protection). For severity measurements prefer triaxial accelerometers on bearing housings; for route-based inspections single-axis units on bearings often suffice. Wireless sensors with onboard processing (edge FFT/envelope) reduce wiring and transmit spectra or condensed features to PLC/SCADA/CMMS while allowing threshold-triggered raw data export for expert analysis [2][4].

Mounting best practices:

• Mount accelerometers directly to bearing housings and structural points with consistent orientation.
• Use threaded studs or magnetic bases depending on permanence; avoid adhesives for repeatable, long-term baselines unless mechanically robust.
• Record mounting location diagrams and sensor IDs for repeatability of route-based measurements [3][5].

3. Data Acquisition and Signal Conditioning

Define acquisition settings per asset: sampling rate, record length, windowing function, averaging and anti-alias filtering. Typical settings:

• For overall spectral analysis: sampling rates from 2–10 kHz with record lengths 1–10 seconds depending on rotational speed.
• For bearing envelope: higher sampling rates (≥20 kHz) and demodulation filters centered on machine resonance bands (typically 5–30 kHz) [4][5].
• Use Hanning or flat-top windows for FFTs to control spectral leakage, and apply synchronous averaging when repetitive signals are present.

4. Baseline Collection and Trend Establishment

Collect baseline data under known healthy operating conditions and store spectra, time records and operating metadata (RPM, load, temperature). Baselines require consistent operating points—measure at the same speed and load for valid trend comparison. Track statistical features (RMS, peak, crest factor, kurtosis) and spectral line amplitudes over time to detect new peaks, amplitude growth, or frequency shifts [1][3][5].

5. Automated Detection and Alerting

Implement a hybrid detection strategy:

• Edge thresholds: configure sensors or gateways to generate alarms on exceedance (e.g., RMS velocity thresholds from ISO 10816 or site-specific limits).
• Spectrum-based alerts: detect new spectral lines or increases at RPM-related frequencies and envelope bands to flag potential bearing issues earlier than overall RMS alerts.
• Machine learning / AI: use classification models to prioritize alarms, but always validate AI outputs with technician inspection and baseline comparisons [2][1].

6. Integration and Workflows

Integrate vibration data feeds into PLC/SCADA for operational context and into CMMS for work-order automation. Map vibration alarms to corrective procedures (inspect, balance, align, replace bearing) and link to historical work records for ROI tracking. Follow ISA-95 principles to connect enterprise maintenance systems with control-level monitoring if required [2].

7. Validation and Continuous Improvement

Validate predictions by correlating spectral findings with physical inspections, oil analysis, and thermography. When false positives occur, tune thresholds, improve mounting repeatability, or refine classifier models. Maintain an audit trail of sensor calibrations and periodic on-site checks to preserve data quality [3][5].

Best Practices

Based on decades of field experience and vendor guidance, follow these proven best practices to achieve reliable PdM results.

• Consistent Measurement Conditions: Use identical sensor locations, mounting method and operating points for both baselines and periodic readings. Variation in load or speed invalidates many comparison metrics [1][3].
• Prioritize High-Impact Assets: Deploy permanent wireless monitors on the top 10–20% of assets that create 80% of downtime risk; use route-based collection on lower-priority equipment [2].
• Combine Continuous and Route-Based Monitoring: Continuous (permanent) monitoring provides immediate alerts; periodic route-based inspections capture broader fleets without full instrumentation, reducing cost [2][4].
• Train Teams on Interpretation: Provide technicians with spectral literacy training—understanding 1×, 2×, harmonics, sidebands and envelope outputs reduces escalations and misdiagnoses. Use training kits (e.g., Amatrol) for hands-on practice [6].
• Use Edge Processing to Reduce Data Volume: Perform FFT/envelope on the sensor or gateway and transmit condensed features/spectra; export raw time data only on alarms for expert analysis [4][2].
• Document and Automate: Maintain clear asset metadata (bearing types, gear tooth counts), alarm thresholds and corrective actions in CMMS to shorten repair cycles and measure program ROI [3].

Signal Processing and Measurement Hygiene

Implement proper signal processing techniques to improve diagnosis accuracy:

• Apply anti-alias filters before digitization and choose sampling rates at least 2.5× the highest frequency of interest.
• Use windowing (Hanning for general use, flat-top for amplitude accuracy) and average multiple records to improve signal-to-noise ratio.
• Monitor time-domain kurtosis and crest factor; sudden increases often precede RMS amplitude changes and indicate impact-type faults [5][8].

Standards, Severity Criteria and Example Limits

Vibration severity evaluation commonly references the ISO 10816 series (now updated in parts and sometimes referenced together with ISO 20816); the series provides guidance for evaluating machine vibration in the time domain and classifying severity into zones. ISO 10816 establishes four condition zones:

Zone	Condition	Typical Action
Zone A	Good / Newly commissioned	Normal operation, continue monitoring
Zone B	Satisfactory	Acceptable operation, plan inspection
Zone C	Unsatisfactory	Investigate and repair
Zone D	Unacceptable	Immediate shutdown and repair

ISO 10816-3 (machines 15–600 kW) gives specific RMS velocity thresholds by speed band. As an illustrative example from public guidance, for machines operating 600–1,800 RPM a value below approximately 2.3 mm/s RMS can be classified as good condition for certain power ranges; however, exact limits vary by machine class and the complete numerical tables reside in the official ISO documents — always consult the published standard for compliance and design decisions [1][5].

Comparison: Typical Sensor and Solution Options

Below is an abbreviated comparison of representative commercial solutions described in industry literature and vendor materials. Verify current product versions and detailed specifications on manufacturer sites before procurement.

Product / Type	Processing	Interface / Integration	Primary Use Case	Notes
GraceSense™ (Graceport)	Onboard FFT and classification	Wireless to IoT gateway, PLC/SCADA/CMMS integration	Permanent monitoring for motors, pumps	AI analytics and defect classification; good for immediate alerts and trending [2]
NCD Vibration Sensor Gen 4 (NCD.io)	Internal FFT, thresholding, raw export on alarm	Wireless gateway / IoT integration	Edge spectra for PdM platforms and gateways	Configurable for tradeoff between edge processing and raw data export [4]
Amatrol 97-ME5A Training System	Educational instrumentation (meter + sensors)	Training lab integration	Hands-on training for vibration diagnosis	Includes motor, belt/shaft drives and vibration meter for training courses [6]

Diagnostics and Fault Mapping

Apply RPM-referenced libraries and bearing geometry to map detected spectral peaks to component fault frequencies:

• Calculate characteristic frequencies (BPFO, BPFI, BSF, FTF) from bearing inner/outer race dimensions and element counts; compare to envelope-demodulated spectra for early bearing defects [8].
• Track harmonics and sidebands around 1× to assess misalignment, belt shear or looseness; sideband spacing at shaft frequency often indicates modulation by rotational speed (e.g., gear mesh sidebands caused by eccentricity).
• Use order analysis for variable-speed machines to map vibrations to rotational orders rather than absolute frequency, improving fault correlation across speed changes [