What is FFT and how to interpret it in industrial vibration analysis

Understanding what FFT (Fast Fourier Transform) is becomes essential for predictive maintenance and reliability professionals who need to correctly interpret the vibration signals of their equipment.

In essence, FFT is a mathematical technique that transforms time‑domain data into frequency‑domain information, revealing details about the dynamic behavior of motors, pumps, gearboxes, and other industrial assets.

In practice, FFT in predictive maintenance is what makes it possible to identify early‑stage failures—such as unbalance, misalignment, and bearing defects—before they turn into unplanned downtime.

By analyzing the frequency spectrum, engineers can visualize peaks and harmonics that indicate the origin of the problem, enabling a faster and more accurate diagnosis based on real data collected by IoT vibration sensors.

In this article, you will understand what FFT is, how it works, and how it can be applied to industrial vibration analysis. We will also explain how to interpret FFT graphs, which parameters influence the reading, and how Dynamox uses its sensors and the Dynamox Platform to automate diagnostics and increase operational reliability.

What is FFT (Fast Fourier Transform)?

To understand what FFT is, it’s important to remember that every piece of equipment in operation generates vibrations—and these vibrations contain valuable information about the machine’s mechanical condition.

The Fast Fourier Transform (FFT) is a mathematical algorithm that allows a complex signal—such as the vibration signal of a vibrating screen—to be decomposed into its constituent frequencies, revealing information that is otherwise hidden within the raw signal.

In other words, what was once just a time‑domain vibration waveform can now be visualized as a frequency spectrum, making it possible to identify which components (bearings, gears, shafts) are contributing to the overall vibration.

FFT is therefore a practical and efficient way of applying the Fourier Transform, a fundamental technique in engineering and physics for signal analysis. While the original transform requires high computational power, the “fast” version simplifies the calculation, enabling sensors and digital platforms to process vibration data in short time intervals. This efficiency is what makes continuous monitoring of industrial assets possible, supporting predictive maintenance based on reliable, real‑time data.

FFT in practice

To understand what FFT is in practice, it is essential to differentiate the two types of signal representation:

• Time domain: shows how the signal varies over time. It is the raw form captured by the sensor, displaying amplitude changes, peaks, and oscillations.
• Frequency domain: shows how the signal’s energy is distributed across different frequencies. Each peak in the FFT plot corresponds to a vibration source associated with a specific machine component (such as the shaft, bearing, or gear).

Therefore, by converting the signal from the time domain to the frequency domain, the FFT reveals behavioral patterns that would otherwise be invisible to the naked eye. For example, an increase in amplitude within a specific frequency range may indicate unbalance, while multiple harmonics may suggest misalignment or mechanical looseness.

Why is FFT essential in vibration‑based maintenance?

In predictive maintenance, FFT is the foundation of mechanical fault diagnosis. It allows early identification of degradation symptoms before a failure occurs, enabling engineers to plan interventions in advance.

Without FFT, vibration signals would be nothing more than complex, hard‑to‑interpret oscillations. With FFT, however, it becomes possible to associate specific frequencies with known failure modes, such as:

• Unbalance: a peak at the shaft rotational frequency.
• Misalignment: presence of multiple harmonics of the fundamental frequency.
• Mechanical looseness: irregular amplitude and scattered harmonics.
• Bearing defects: peaks at specific characteristic frequencies (BPFO, BPFI, BSF, FTF).

Therefore, understanding what FFT is means understanding the core of vibration analysis. This technique transforms raw data into valuable insights, enabling assertive decisions, fewer unplanned shutdowns, and increased operational reliability.

How is FFT applied in predictive maintenance?

Understanding what FFT is and how to apply it is the first step toward transforming vibration data into high‑precision diagnostics. In predictive maintenance, FFT converts the signals collected by vibration sensors, such as DynaLoggers, into useful information. This information reveals abnormal operating conditions before a failure becomes critical.

Role of FFT in vibration analysis

The main function of FFT is to isolate the vibration frequencies present in a piece of equipment, allowing them to be correlated with its mechanical components. Each part of the machine — shaft, bearing, gear, coupling — vibrates at a characteristic frequency.

Thus, when an anomaly occurs, such as wear, unbalance, or looseness, that frequency changes, and the FFT spectrum displays a specific peak or additional harmonics, enabling the analyst to identify the source of the problem.

In this sense, FFT works like an “X‑ray” of the asset’s dynamic behavior. It provides a vibration signature that can be compared with previous measurements to detect failure trends. This continuous analysis supports condition‑based maintenance, one of the most advanced stages of the predictive maintenance journey.

Difference between the time‑domain signal and the frequency spectrum

The signal in the time domain is the vibration recorded directly by the sensor, reflecting the raw motion. This signal is actually a combination of several simpler components (or signal frequencies), each related to different parts or phenomena within the machine. However, when analyzing only the time‑domain signal, these components appear overlapped, making individual identification and interpretation difficult.

By applying FFT, the signal is converted to the frequency domain, where each frequency appears isolated. This allows the analyst to visualize its intensity and identify its origin, making failure analysis much more precise and efficient.

Thus, while the time‑domain signal shows when a vibration occurs, the frequency spectrum reveals why it occurs, indicating the physical source of the problem.a. This is the fundamental difference between observing the apparent behavior and understanding the root cause of the failure.

Therefore, FFT is considered the link between raw data and technical decision‑making. It translates the machine’s behavior into a format that engineers can interpret, correlate, and use to plan interventions in advance, reducing costs and preventing unplanned downtime.

How to interpret an FFT plot

Understanding what FFT is remains essential, but the true value of the technique lies in interpreting the frequency spectrum. The FFT plot shows how vibrational energy is distributed across different frequencies, allowing the identification of mechanical fault patterns based on peaks, harmonics, and noise.

Meaning of the axes: frequency and amplitude

An FFT plot is composed of two main axes:

• Frequency (Hz): indicates how many times the vibratory motion repeats per second. Each machine component vibrates at a characteristic frequency.
• Amplitude: indicates the vibration intensity at each frequency. It can be displayed in acceleration (g, m/s²), velocity (mm/s), or displacement (µm). The higher the peak, the greater the amount of energy associated with that vibration.

Each unit highlights a different physical phenomenon:

• High frequencies are best observed in acceleration;
• Moderate faults and dynamic balance issues are best observed in velocity;
• Very low frequencies are best observed in displacement.

What are harmonics, peaks, and noise?

Harmonics, peaks, and noise make up the “vocabulary” of the spectrum. Therefore, understanding their physical origin is key to associating them with failure causes.

• Fundamental peaks: correspond to the main excitation frequency (e.g., 1× RPM). Their increase in amplitude typically indicates abnormal conditions related to the same excitation mechanism.
• Harmonics (n× fundamental): integer multiples of the main frequency. They indicate mechanical nonlinearities and asymmetries, commonly seen in misalignment and looseness, for example. Additionally, the amplitude ratio between harmonics helps classify the type of defect.
• Sidebands: peaks regularly spaced around a given frequency (e.g., gear mesh). They suggest modulation caused by load variation, eccentricity, or periodic defects. When spaced by the rotational frequency, they are characteristic of mechanical modulation.
• Background noise: a diffuse rise in spectral level without defined peaks. It may result from turbulence, general friction, electrical interference, or structural coupling. Thus, an increase in high‑frequency noise may indicate early‑stage bearing defects. This sensitivity becomes even greater when analyzed using demodulation techniques (envelope analysis).

Parameters that influence FFT analysis

The quality of vibration analysis depends directly on how the signal is acquired and processed. Therefore, even when the analyst understands what FFT is and how to interpret the spectrum, errors in configuring measurement parameters can distort the results and lead to incorrect diagnostics.

Among the most important factors are the sampling rate, spectral resolution, the type of window applied to the signal, and the total acquisition time. Each of these elements defines the level of detail and fidelity of the information obtained. See below:

Sampling rate, resolution, and number of lines

The sampling rate determines how many data points per second the system collects from the vibration signal. According to the Nyquist Theorem, the maximum frequency (Fmax) that can be analyzed is equal to half of the sampling rate. Therefore, investigations of high‑frequency defects — such as early‑stage bearing faults — require high sampling rates to ensure adequate coverage of the frequency range of interest.

Spectral resolution depends on the number of analysis lines (N) and the configured Fmax. The relationship between them is:

A higher resolution (more lines) allows very close peaks to be distinguished, which is essential for separating, for example, the shaft rotational frequency from the harmonic frequencies of a gear.

On the other hand, excessively high resolutions increase processing time and data volume, so teams must balance these parameters according to the type of asset being monitored.

Best practices for spectral analysis

In practical industrial measurements, many analysts use 1,600 lines of resolution as a starting point for global monitoring. For equipment with richer high‑frequency content, it is common to increase to 3,200, 6,400, or more lines to capture finer details at higher frequencies. In addition, the ideal number depends on the maximum frequency range (Fmax), the criticality of the asset, and the capability of the acquisition system. See the table below:

In this scenario of advanced diagnostics, Dynamox’s wireless monitoring solutions, such as the DynaLogger HF+, increase analytical capacity by enabling ultra‑high‑fidelity configuration options.

Our sensors are capable of collecting data and generating spectra with up to 98,304 lines of resolution (LOR) in uniaxial measurements and 32,768 lines in triaxial measurements.. This processing power ensures the spectral resolution necessary to distinguish even the faintest harmonic frequencies in complex bearing and gear defects, meeting the needs of the most critical and demanding assets.

Window types: Hanning, Hamming, and Flat Top

During acquisition, the vibration signal rarely contains an exact integer number of cycles. To prevent discontinuities between the start and end of the sample — which cause a phenomenon known as spectral leakage — a weighting window is applied to smooth the edges of the signal before computing the FFT.

• Hanning window: the most commonly used in vibration measurements. It offers a good balance between frequency resolution and amplitude accuracy, making it suitable for general diagnostics.
• Hamming window: similar to the Hanning window but with slightly lower sidelobe attenuation. The Hamming window reduces leakage more aggressively, though with a small loss in the ability to separate very closely spaced frequencies. It is used when minimizing discontinuities is a priority, even at the cost of fine‑resolution detail.
• Flat Top window: ideal when the priority is to measure absolute amplitude with high accuracy, even if resolution is sacrificed. It is recommended for calibrations and comparative testing.

Thus, the choice of window directly influences the clarity and reliability of the FFT spectrum. In industrial measurements, the Hanning window is generally the standard recommended by the ISO 10816 and ISO 20816 norms.

Acquisition time and precautions to avoid aliasing

The total acquisition time determines the interval over which the FFT will be calculated. Very short sampling durations reduce resolution, while long recordings may include transient variations that distort the spectrum.

Therefore, the ideal approach is to select a duration that captures complete rotation cycles of the asset, ensuring stability in the signal.

Another crucial precaution is avoiding aliasing — an error that occurs when the sampling rate is insufficient to correctly represent the signal’s frequencies. In such cases, false peaks appear in incorrect regions of the spectrum, compromising the analysis.

• Practical recommendation: in vibration measurements, set the sampling rate with an adequate safety margin relative to the maximum frequency of interest (Fmax). This ensures the effective performance of the anti‑aliasing filter. In this way, the FFT spectrum accurately represents the signal’s behavior. Technically, Fmax must remain below half of the sampling rate — known as the Nyquist frequency — to avoid aliasing and spectral distortion.

In the case of DynaLogger sensors, this risk is eliminated by anti‑aliasing filters that ensure the Fmax of interest remains within the Nyquist frequency. This guarantees that the FFT spectrum reflects only the machine’s actual vibration, protecting the analysis from artifacts and distortions.

In the end, understanding what FFT is also means understanding how to configure it correctly. Proper selection of sampling rate, window type, and acquisition time ensures reliable spectra and accurate diagnostics, reducing analytical noise and increasing sensitivity to early‑stage fault detection.

How Dynamox uses vibration analysis to support predictive diagnostics

Dynamox provides a complete condition‑monitoring ecosystem for predictive maintenance, combining IoT sensors, gateways, analytical software, and artificial intelligence. This integration enables vibration and temperature data to be transformed into practical information for fault diagnosis and asset management.

The DynaLogger sensors continuously monitor vibration and temperature, recording trends and frequency spectra that reflect the dynamic behavior of assets. Then, the DynaGateways automatically send the data to the Dynamox Platform, which organizes the measurements into intuitive dashboards and reports, making it easier to visualize patterns and anomalies.

The artificial intelligence module, DynaDetect, complements the process by analyzing the collected signals and issuing automatic alerts for potential mechanical failures such as unbalance, misalignment, or bearing defects. This automation speeds up technical interpretation, increases diagnostic reliability, and reduces maintenance team response time.

Together, these tools make monitoring more efficient, standardized, and data‑driven, contributing to greater operational availability and asset reliability across industrial environments.

To learn more, talk to a Dynamox specialist and discover how to apply vibration analysis, smart sensors, and artificial intelligence to increase reliability and reduce maintenance costs in your industry.

Success Story

The accuracy of FFT parameters is essential in critical assets, such as low‑speed belt conveyors. In a successful case at Ferro+ Mining, the DynaLoggers and the Dynamox Platform demonstrated the value of high spectral resolution.

The team was able to detect and isolate the exact frequency of an outer race bearing fault (BPFO), even with the machine operating at only 60 RPM.

At low speeds, the characteristic amplitudes of bearing defects naturally appear very low in the raw spectrum, making it essential to use high‑fidelity signals and high‑resolution processing. The precise diagnosis allowed the mining company to plan the component replacement, preventing a catastrophic failure and avoiding unplanned downtime.

Discover how Dynamox solutions transform raw data into highly reliable predictive insights. Click to read the full success story.

FAQ – Frequently Asked Questions about FFT