Calculating the RMS Energy of the FFT

Jack D. Peters
Accelent Technology LLC
19 Olde Harbour Trail
Rochester, New York, 14612
[email protected]

Abstract: The Fast Fourier Transform (FFT) is used in vibration analysis. It is well known that
Window Filters (Hanning, Flat Top, etc.) can be applied to the time waveform during the batch
processing of the FFT, and as a result, decrease the resolution of the FFT. The resolution is
decreased by a numerical Window Factor based on the applied filter. It is not well known that the
Window Factor (WF) can also be described as the Equivalent Noise Bandwidth (ENBW) or Noise
Power Bandwidth. This paper will explain how to calculate the RMS Energy of the FFT when the
WF or ENBW used to process the FFT has a numerical value greater than one.

Key Words: Time Waveform, Fast Fourier Transform, FFT, Window Factor, Window Filter,
Equivalent Noise Bandwidth, Noise Power Bandwidth, RMS Energy, RMS Power, Lowest
Resolvable Frequency, Vibration Analysis

Introduction:
The Fast Fourier Transform or FFT was developed by James Cooley (IBM) and John
Tukey (Princeton) in 1965. The FFT included the introduction of computer processing of the
Fourier Series originally developed by a French mathematician and physicist, Joseph Fourier. This
highly improved the accuracy of vibration analysis for resolving closely spaced frequencies in the
spectrum. [1]
In the form of an algorithm or calculation, the FFT is a batch process that is applied to
individual blocks of time waveform data that typically represent the mechanical vibration of the
machine as measured by an accelerometer, velocity sensor or displacement probe. Figure 1
illustrates a typical time waveform and FFT.

©Jack D. Peters, 27 April 2020 1 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Figure 1 Time Waveform (upper trace) & FFT (lower trace)

The FFT shown in Figure #1, has a vertical axis in acceleration (g’s RMS) and a horizontal
axis in frequency (Hz). The individual frequencies identified in the FFT each have an amplitude
value. The highest amplitude, 0.0465 g’s RMS, is identified at 72.5 Hz. The amplitudes of the
individual frequencies in the FFT are important in assessing the condition of the machine (i.e.
Unbalance, Misalignment or Looseness) or the condition of individual components in the machine
(i.e. Rolling Element Bearings or Gears). Although the amplitudes of the individual frequencies in
the FFT could be trended for a change in amplitude indicating a change in condition, it would
require a significant amount of processing and data storage. This may not provide an efficient
method for general evaluation of the machine. The Overall RMS Value of the FFT describes the
RMS energy contained in the FFT. In Figure 1, the Overall RMS Value is 0.0602 g’s RMS and it
describes the RMS Energy contained in the FFT from 0 – 1000 Hz. This overall value of the RMS
Energy is easy to trend and alarm. If the energy increase or decreases, it indicates a change in
condition. In some applications, it may be more appropriate to measure the RMS Band Energy to
trend known changes in specific machine conditions or components. The RMS Band Energy is
calculated for a range of frequencies instead of the total frequency span of the FFT. The RMS
Band Energy from 50 – 100 Hz would specifically track the amplitude changes in a range that
includes the 72.5 Hz, in Figure #1, but ignores all other amplitudes.

©Jack D. Peters, 27 April 2020 2 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Bins and Lines:
The horizontal axis on the FFT is defined as the frequency axis. Typical units will be Hz
(Hertz), CPS (cycles/second) or CPM (cycles/minute). When a frequency is displayed on the
horizontal axis, it is presented as a Line that represents all of the frequencies in the Bin. The number
of Lines in the FFT is a user selectable feature of the data collector, vibration analyzer or dynamic
signal analyzer being used in the measurement process. User selectable Lines typically include;
100, 200, 400, 800, 1600, 3200, 6400, 12800 and 25600. Examples of Bin Width are shown:

Example 1: Frequency span is 0 – 500 Hz, Number of lines selected are 400

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛 500 𝐻𝑧 1.25 𝐻𝑧

𝐵𝑖𝑛 𝑊𝑖𝑑𝑡ℎ = = =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠 400 𝐿𝑖𝑛𝑒𝑠 𝐿𝑖𝑛𝑒

Example 2: Frequency span is 0 – 500 Hz, Number of lines selected are 1600

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛 500 𝐻𝑧 0.3125 𝐻𝑧

𝐵𝑖𝑛 𝑊𝑖𝑑𝑡ℎ = = =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠 1600 𝐿𝑖𝑛𝑒𝑠 𝐿𝑖𝑛𝑒

As indicated by Examples 1 and 2, if the number of lines is increased and the frequency
span remains constant, the Bin width is decreased. It would indicate that a smaller Bin width would
have better resolution. In fact, the Bin width describes the Lowest Resolvable Frequency (LRF)
that can be identified in the FFT display. Given the 0 – 500 Hz frequency span and 400 Lines of
Example #1, identifiable frequencies in the FFT would be; 0, 1.25, 2.50, 3.75 - - - 496.00, 497.25,
498.5 and 500 Hz. The cursor in the FFT display could be successfully moved from Line to Line
indicating the frequency and amplitude of each Line. As shown in Figures 2, 3 and 4, the Line is
always in the center of the Bin width. A frequency of 35 Hz has been used to illustrate the Bin
width, and reflects the calculations of Examples 1 and 2.

Lines 28th Line = 35.00 Hz 112th Line = 35.00 Hz

Amplitude

Amplitude
Amplitude

Bins 34.375 Hz 35.625 Hz 34.84375 35.15625

Hz Hz
Frequency Frequency Frequency
Figure 2 Bins & Lines Figure 3 1.25 Hz/Line Figure 4 0.3125 Hz/Line

©Jack D. Peters, 27 April 2020 3 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Window Filters:
The FFT algorithm requires continuous data for calculation and processing. If the data in
the time waveform is discontinuous, the frequencies displayed on the horizontal axis of the FFT
will be poorly defined or as it is commonly described, they will have leakage. In vibration analysis,
the FFT is typically a batch process using multiple samples of time waveform data, often referred
to as averages or time blocks. Although each time block contains a sample of the time waveform
data, the data may not be periodic and when the time blocks are averaged to calculate the FFT, the
data is discontinuous. To correct the discontinuity between the individual time blocks a Window
Filter can be applied. The Window Filter will modify the time waveform so that it starts and stops
at a zero amplitude in each individual time block forcing the data to be periodic. In this manner,
when the time blocks are averaged, they simulate a continuous time waveform and conform to the
requirement of the FFT algorithm. There are multiple Window Filters available to the user. The
most popular window filters for vibration analysis are Hanning and Flat Top. An illustration of the
Window Filter effect on the time waveform is shown in Figures 5, 6 and 7.

Figure 5 No Window Filter, Non-Periodic, Discontinuous Time Waveform Data

Figure 6 Hanning Window Filter, Periodic, Continuous Time Waveform Data

Figure 7 Flat Top Window Filter, Periodic, Continuous Time Waveform Data

©Jack D. Peters, 27 April 2020 4 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Window Filters & Resolution:
Applying the Hanning or Flat Top Window Filter to the time waveform data will decrease
the frequency resolution of the FFT. This means that closely spaced frequencies will not have
enough resolution to separate them and they will appear as one frequency.
Window filters have an associated Window Factor (WF) with them. This Window Factor
is a numerical value describing the Equivalent Noise Bandwidth [2] or Noise Power Bandwidth
[3]. Window Factors (WF) or Equivalent Noise Bandwidths (ENBW) are shown in Table 1.

Window Filter WF or ENBW

Uniform 1.00
Rectangular 1.00
Hanning 1.50
Flat Top 2.96
Flat Top 3.82
Table 1 Window Factors or Equivalent Noise Bandwidth
It should be noted in Table 1 that some Window Filters will have multiple values for the
ENBW based on the calculated design of the Window Filter by the manufacturer of the data
collector, vibration analyzer or dynamic signal analyzer. This author has two vibration analyzers.
Both of them have an ENBW of 1.5 for the Hanning Window Filter, but for the Flat Top Window
Filter, one of them has an ENBW of 2.96 and the other one has an ENBW of 3.82. The user should
always refer to the technical specifications for their analyzer or ask the manufacturer to ascertain
the ENBW of the Window Filter they are using.

In Examples 1 and 2, the Bin Width or Lowest Resolvable Frequency (LRF) was identified
by dividing the Frequency Span by the Number of Lines. When a Window Filter is used, the
Window Factor or ENBW must be combined with the LRF. This combination will identify the
Bandwidth [4] of the FFT, as shown in Examples 3 and 4.

Example 3: Frequency span is 0 – 500 Hz, Number of lines are 400, Hanning Window

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛 500 𝐻𝑧 1.875 𝐻𝑧

𝐵𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ = 𝐸𝑁𝐵𝑊 = 1.5 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠 400 𝐿𝑖𝑛𝑒𝑠 𝐿𝑖𝑛𝑒

Example 4: Frequency span is 0 – 500 Hz, Number of lines are 1600, Hanning Window

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛 500 𝐻𝑧 0.46875 𝐻𝑧

𝐵𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ = 𝐸𝑁𝐵𝑊 = 1.5 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠 1600 𝐿𝑖𝑛𝑒𝑠 𝐿𝑖𝑛𝑒

©Jack D. Peters, 27 April 2020 5 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 In January of 1978, Jack Frarey demonstrated that the Bandwidth was not sufficient to
separate two closely spaced frequencies [5]. Mr. Frarey’s work indicated that the true resolution
of the FFT was calculated by multiplying the Bandwidth by a factor of two. The calculated
Resolution of the FFT is shown in Examples 5 and 6.

Example 5: Frequency span is 0 – 500 Hz, Number of lines are 400, Hanning Window

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛 500 𝐻𝑧 3.75 𝐻𝑧

𝑅𝑒𝑠𝑜𝑙𝑢𝑡𝑖𝑜𝑛 = 2 𝐸𝑁𝐵𝑊 = 2 1.5 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠 400 𝐿𝑖𝑛𝑒𝑠 𝐿𝑖𝑛𝑒

Example 6: Frequency span is 0 – 500 Hz, Number of lines are 1600, Hanning Window

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛 500 𝐻𝑧 0.9375 𝐻𝑧

𝑅𝑒𝑠𝑜𝑙𝑢𝑡𝑖𝑜𝑛 = 2 𝐸𝑁𝐵𝑊 = 2 1.5 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠 1600 𝐿𝑖𝑛𝑒𝑠 𝐿𝑖𝑛𝑒

Window Filters & Scallop Loss:

Applying the Hanning or Flat Top window filter to the time waveform data can affect the
amplitude of the frequency components in the FFT. Figures 8, 9, 10 and 11 illustrate these two
Window Filters and the Scallop Loss effect on the peak amplitude of the FFT based on a frequency
span of 400 Hz with 400 lines (400 Hz/400 Line = 1 Hz/Line) or a 1 Hz Bin Width [2 & 6].

Scallop Loss
Hanning -15.1 % (1.42 dB)
100
90
80
70
60
50
%

40
30
20
10
0
0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz)
A B

Figure 8 Hanning Window Filter (ENBW = 1.5) with Scallop Loss

©Jack D. Peters, 27 April 2020 6 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Scallop Loss
Flat Top
-1.0 % (0.1 dB)
100
98
96
94
92
90
%

88
86
84
82
80
0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz)
A B

Figure 9 Flat Top Window Filter (ENBW = 3.8) with Scallop Loss

The narrow Bandwidth of the Hanning Window Filter in Figure 8 provides good frequency
resolution for general purpose vibration measurements but has a high amplitude uncertainty.
Frequency A in Figure 8 is centered in the filter band and will have an accurate amplitude value.
Frequency B in Figure 8 is at the intersection of two filter bands and will have an amplitude value
that is attenuated due to Scallop Loss by 15.1%.
The wider Bandwidth of the Flat Top Window Filter in Figure 9 provides poor frequency
resolution for general purpose vibration measurements, but has a low amplitude uncertainty.
Frequency A in Figure 9 is centered in the filter band and will have an accurate amplitude value.
Frequency B in Figure 9 is at the intersection of two filter bands and will have an amplitude value
that is attenuated due to Scallop Loss by only 1%.
In Figure 10, the Hanning Window Filter (ENBW = 1.5) and Flat Top Window Filter
(ENBW = 3.8) have been superimposed on each other. Two examples are shown. One is bin
centered at 3 Hz and the other is bin centered at 7 Hz. Vertical scaling is 0% to 120% so that both
filter shapes are fully visible.
 The Bandwidth of the Hanning Filter is equal to (400 Hz/400 Lines) 1.5 = 1.5 Hz.
 The Bandwidth of the Flat Top Filter is equal to (400 Hz/400 Lines) 3.8 = 3.8 Hz.
In Figure 11, vertical scaling has been adjusted and is set at 80% to 100%. This provides
an expanded view and comparison of the Bin Width and Scallop Loss of each filter.
Although Scallop Loss was an important aspect of window filters when they were first
applied to digital signal analysis, many of today’s vibration analyzers, data collectors and virtual
instruments are able to measure and display the correct peak value with no user interaction.

©Jack D. Peters, 27 April 2020 7 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Hanning & Flat Top
120

100

80

60
%

40

20
1.5 Hz 3.8 Hz
0
0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz)

Figure 10 Hanning & Flat Top Window Filters Bin Centered at 3 Hz and 7 Hz

Hanning & Flat Top

105
Bin Bin
-1%
100
Hanning
95
-15.1%
%

90
Flat Top
85

80
0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz)

Figure 11 Hanning & Flat Top Window Filters with Bin Width & Scallop Loss

©Jack D. Peters, 27 April 2020 8 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Window Filters and RMS Energy:
Applying the Hanning or Flat Top Window Filter to the time waveform data will affect the
RMS Energy of the FFT. In Figure 12, the fundamental frequency is 72.5 Hz and there are ten
additional orders with visible amplitudes (g’s RMS). The vibration analyzer has calculated the
overall RMS Energy at 0.0602 g’s RMS.

0.047

0.020
0.011 0.013
0.007 0.008 0.009
0.004 0.006 0.005
0.003

Figure 12 FFT with Multiple Orders of the Fundamental Frequency

An approximation of the overall RMS Energy could be made by using the individual
amplitude of the fundamental frequency and amplitude of each order. This calculation could be
made by; square root sum of squares, as shown in Example 7:

Example 7: Square Root Sum of Squares

𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 = . 047 +. 007 +. 008 +. 020 +. 004 +. 011 +. 006 +. 013 +. 003 +. 005 +. 009

𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 =0.0564 g’s RMS

The approximation in Example 7 has not used all of the energy that is contained in the FFT
from 0 – 1000 Hz. It has only calculated an approximation of eleven amplitudes. To correctly
calculate the overall RMS Energy of the FFT the equation in Example 8 should be used.

Example 8:

𝐴 +𝐴 +𝐴 +⋯𝐴 𝐴 +𝐴 +𝐴 +⋯𝐴
𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 = 𝑜𝑟 𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 =
𝐸𝑁𝐵𝑊 𝑊𝐹

Where:

©Jack D. Peters, 27 April 2020 9 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 𝐴 = 𝑇ℎ𝑒 𝐴𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 𝑜𝑓 𝑒𝑎𝑐ℎ 𝐿𝑖𝑛𝑒 𝑖𝑛 𝑡ℎ𝑒 𝐹𝐹𝑇 (𝑔 𝑠 𝑅𝑀𝑆 𝑜𝑟 𝑖𝑛 𝑠 𝑅𝑀𝑆 𝑜𝑟 𝑚𝑚⁄𝑠 𝑅𝑀𝑆)

𝑁 = 𝑇ℎ𝑒 𝑙𝑎𝑠𝑡 𝐿𝑖𝑛𝑒 𝑖𝑛 𝑡ℎ𝑒 𝐹𝐹𝑇

𝐸𝑁𝐵𝑊 = 𝐸𝑞𝑢𝑖𝑣𝑎𝑙𝑒𝑛𝑡 𝑁𝑜𝑖𝑠𝑒 𝐵𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ

𝑊𝐹 = 𝑊𝑖𝑛𝑑𝑜𝑤 𝐹𝑎𝑐𝑡𝑜𝑟

The FFT shown in Figure 12 utilized a Hanning Window Filter and 400 Lines. The ENBW
or WF will be 1.5. To calculate the equation in Example 8, it would be necessary to export the
frequency and amplitude data from the FFT in Figure 10 and import it to Excel or some other type
of software that can process a large volume of data. Calculations for 400 Lines with amplitudes
from 0 – 1000 Hz will be required. Table 2 is an example on how this could be done in Excel and
contains 400 amplitudes from 0 – 1000 Hz. Each amplitude (g’s RMS) is in column 2 and has been
squared in column 3 (g’s2 RMS). The sum of squares has been calculated as 0.00543711 g’s 2 RMS.
This sum will be divided by the ENBW or WF, and since we are using a Hanning Window Filter,
the divisor will be 1.5 providing a result of
0.00362474 g’s2 RMS. The square root of this Hz g’s RMS g's2 RMS
value is 0.06020583 g’s RMS and represents the 0.0 0.00724940 0.00005255
overall energy of the FFT as calculated by the 2.5 0.00362080 0.00001311
vibration analyzer. 5.0 0.00012540 0.00000002
(To reduce the size of Table 2, values between 10 7.5 0.00024780 0.00000006
Hz and 990 Hz are not shown, but they have been 10.0 0.00021800 0.00000005
included in the calculations.) - - -
- - -
- - -
990.0 0.00007931 0.00000001
If the square root sum of squares had been
992.5 0.00011210 0.00000001
used to calculate the overall RMS Energy from 0 –
995.0 0.00016160 0.00000003
1000 Hz in Figure 12, the value of the RMS Energy
997.5 0.00019500 0.00000004
would have been too high as shown in Example 9.
1000.0 0.00030960 0.00000010
The value calculated by using the square Sum of Squares 0 – 1000 Hz 0.00543711
root sum of squares is approximately 22% too high Divide by 1.5 0.00362474
when compared to the actual value that includes the Calculate Square Root 0.06020583
influence of the ENBW or WF on the RMS Energy. Table 2 RMS Energy Calculation
In order to gain a better understanding of how the
ENBW or WF effects the RMS energy of the FFT, the analysis will be limited to two single
harmonic frequencies, 35 Hz and 122 Hz, with no additional noise or influence from other sources.

Example 9: 𝑆𝑞𝑢𝑎𝑟𝑒 𝑅𝑜𝑜𝑡 𝑆𝑢𝑚 𝑜𝑓 𝑆𝑞𝑢𝑎𝑟𝑒𝑠 = √0.00543711 = 0.07373676 𝑔 𝑠 𝑅𝑀𝑆

©Jack D. Peters, 27 April 2020 10 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Understanding RMS Energy with a Hanning Window Filter:
If an FFT with 400 Lines, Hanning Window and 500 Hz frequency span is used, the LRF
or Bin width will be 500 Hz/400 Lines = 1.25 Hz/Line. The 35 Hz frequency will be Bin centered
because 35 Hz/(1.25Hz/Line) = 28 Lines. This indicates that the amplitude of 35 Hz frequency
should be measured correctly by the 28th Line. The 122 Hz frequency will not be Bin centered
because 122 Hz/(1.25Hz/Line) = 97.6 Lines. This indicates that the amplitude of the 122 Hz
frequency will be measured by the 97th and 98th Lines of the FFT. This means that there will be
shared amplitudes at 121.25 Hz and 122.50 Hz since the frequency is not Bin centered. The FFT
in Figure 13 represents the 35 Hz and 122 Hz frequencies with a Hanning Window Filter.

35 Hz 122.5 Hz
1.004 g’s RMS 1.006 g’s RMS

Figure 13 35 Hz and 122 Hz Representation in the FFT (Hanning Window)

35 Hz
1.004 g’s RMS

33.75 Hz 36.25 Hz
0.5016 g’s RMS 0.5026 g’s RMS

Figure 14 35 Hz Expanded View (Hanning Window)

In Figure 14, the FFT has been expanded to focus on the Bin centered 35 Hz frequency.
The modulation like effect of the Window Filter has created what appear to be sideband amplitudes
+/- one Bin of center. The ENBW or WF has created amplitudes above and below the 35 Hz
frequency that must be accounted for. The RMS Energy for the 35 Hz frequency, using a Hanning
Window Filter with an ENBW of 1.5 will be:

0.5016 + 1.004 + 0.5026

𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 = = 1.004 𝑔 𝑠 𝑅𝑀𝑆
1.5

©Jack D. Peters, 27 April 2020 11 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 121.25 Hz 122.5 Hz
0.8779 g’s RMS 1.006 g’s RMS
120 Hz 123.75 Hz
0.1342 g’s RMS 0.2527 g’s RMS
118.75 Hz 125 Hz
0.0224 g’s RMS 0.0296 g’s RMS

Figure 15 122 Hz Expanded View (Hanning Window)

In Figure 15, the FFT has been expanded to focus on the 122 Hz frequency that is not Bin
centered. The ENBW or WF has created amplitudes above and below the 122 Hz frequency that
must be accounted for. The RMS Energy for the 122 Hz frequency, using a Hanning Window
Filter with an ENBW of 1.5 will be:

0.0224 + 0.1342 + 0.8779 + 1.006 + 0.2527 + 0.0296

𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 = = 1.115 𝑔 𝑠 𝑅𝑀𝑆
1.5

The calculations for Figures 14 and 15 provide three very important pieces of information
about calculating the RMS Energy in the FFT using a Hanning Window Filter.
1. If a single harmonic frequency is Bin centered (Figure 14) using a Hanning Window
Filter, the RMS Energy will be equal to the RMS amplitude of that Bin centered
frequency.
2. If a single harmonic frequency is not Bin centered (Figure 15) using a Hanning Window
Filter, the RMS Energy must be calculated as shown because the frequency is
distributed in more than one Bin.
3. The 35 Hz and 122 Hz used in Figures 14 and 15 are single harmonic frequencies.
There is no additional noise added from other frequencies that may appear in an FFT
of actual vibration data.

©Jack D. Peters, 27 April 2020 12 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Understanding RMS Energy with a Flat Top Window Filter:
If an FFT with 400 Lines, Flat Top Window and 500 Hz frequency span is used, the LRF
or Bin width will be 500 Hz/400 Lines = 1.25 Hz/Line. The 35 Hz frequency will be Bin centered
because 35 Hz/(1.25Hz/Line) = 28 Lines. This indicates that the amplitude of 35 Hz frequency
should be measured correctly by the 28th Line. The 122 Hz frequency will not be Bin centered
because 122 Hz/(1.25Hz/Line) = 97.6 Lines. This indicates that the amplitude of the 122 Hz
frequency will be measured by the 97th and 98th Lines of the FFT. This means that there will be
shared amplitudes at 121.25 Hz and 122.50 Hz because the frequency is not Bin centered. The FFT
in Figure 16 represents the 35 Hz and 122 Hz frequencies with a Flat Top Window Filter.

35 Hz 122.5 Hz
1.002 g’s RMS 1.115 g’s RMS

Figure 16 35 Hz and 122 Hz Representation in the FFT (Flat Top Window)

35 Hz
1.002 g’s RMS
33.75 Hz 36.25 Hz
0.9285 g’s RMS 0.9289 g’s RMS

32.5 Hz 37.5 Hz
0.3523 g’s RMS 0.3532 g’s RMS

Figure 17 35 Hz Expanded View (Flat Top Window)

In Figure 17, the FFT has been expanded to focus on the Bin centered 35 Hz frequency.
The modulation like effect of the Window Filter has created what appear to be sideband amplitudes
+/- of the center Bin. The ENBW or WF has created amplitudes above and below the 35 Hz
frequency that must be accounted for. The RMS Energy for the 35 Hz frequency, using a Flat Top
Window Filter with an ENBW of 2.96 will be:

0.3523 + 0.9285 + 1.002 + 0.9289 + 0.3532

𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 = = 1.002 𝑔 𝑠 𝑅𝑀𝑆
2.96

©Jack D. Peters, 27 April 2020 13 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 121.25 Hz 122.5 Hz
1.108 g’s RMS 1.115 g’s RMS
120 Hz 123.75 Hz
0.6922 g’s RMS 0.8343 g’s RMS
118.75 Hz 125 Hz
0.0706 g’s RMS 0.1503 g’s RMS

Figure 18 122 Hz Expanded View (Flat Top Window)

In Figure 18, the FFT has been expanded to focus on the 122 Hz frequency that is not Bin
centered. The ENBW or WF has created amplitudes above and below the 122 Hz frequency that
must be accounted for. The RMS Energy for the 122 Hz frequency, using a Flat Top Window
Filter with an ENBW of 2.96 will be:

0.0706 + 0.6922 + 1.108 + 1.115 + 0.8343 + 0.1503

𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 = = 1.114 𝑔 𝑠 𝑅𝑀𝑆
2.96

The calculations for Figures 17 and 18 provide three very important pieces of information
about calculating the RMS Energy in the FFT using a Flat Top Window Filter.
1. If a single harmonic frequency is Bin centered (Figure 17) using a Flat Top Window Filter,
the RMS Energy will be equal to the RMS amplitude of that Bin centered frequency.
2. If a single harmonic frequency is not Bin centered (Figure 18) using a Flat Top Window
Filter, the RMS Energy must be calculated as shown because the frequency is distributed
in more than one Bin, but since the Flat Top Window only has a 1% maximum amplitude
error as previously discussed, the RMS Energy and amplitudes at 121.25 Hz and 122.5 Hz
will be almost the same..
3. The 35 Hz and 122 Hz used in Figures 17 and 18 are single harmonic frequencies. There
is no additional noise added from other frequencies that may appear in an FFT of actual
vibration data.

Summary:
Window Filters are necessary for vibration analysis because they modify the time
waveform data in each time block to simulate continuous data during the batch processing of the
FFT. Modifying the original time waveform data creates many cause and effects in the FFT that
must be understood for measurement and analysis of the machine vibration.

©Jack D. Peters, 27 April 2020 14 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 The Window Factor or Equivalent Noise Bandwidth have been used interchangeably in
this paper. The Window factor is known terminology that has been used with the calculation of
frequency resolution for many years. The frequency resolution of the FFT can be calculated as
shown:
𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛
𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑅𝑒𝑠𝑜𝑙𝑢𝑡𝑖𝑜𝑛 = 2 𝑊𝑖𝑛𝑑𝑜𝑤 𝐹𝑎𝑐𝑡𝑜𝑟
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠

The Window Factor is also known to affect the amplitude uncertainty of the individual
frequencies being measured. If the frequency is not centered in the Window Filter there will be
some amount of amplitude error. In a Hanning Window Filter the amplitude uncertainty can be
-15.1%, but in a Flat Top Window Filter the amplitude uncertainty is only -1%. In general purpose
vibration measurements, the Hanning Window Filter is preferred for better frequency resolution,
but if the amplitude is critical to the measurement, then the Flat Top Window Filter can be used.

The Equivalent Noise Bandwidth is terminology better used to describe the calculation of
the RMS Energy in the FFT. The amplitudes that are present in adjacent Bins of the FFT contribute
to the noise bandwidth and must be accounted for when calculating the RMS Energy in the FFT
as shown:

If FFT amplitudes (A) are in RMS units (i.e. g’s RMS, in/s RMS or mm/s RMS) then:

𝐴 + 𝐴 + 𝐴 + ⋯𝐴
𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 =
𝐸𝑞𝑢𝑖𝑣𝑎𝑙𝑒𝑛𝑡 𝑁𝑜𝑖𝑠𝑒 𝐵𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ

If FFT amplitudes (A) are in Peak units (i.e. g’s Peak, in/s Peak or mm/s Peak) then:

𝐴 + 𝐴 + 𝐴 + ⋯𝐴
𝑅𝑀𝑆 𝐸𝑛𝑒𝑟𝑔𝑦 = 0.707
𝐸𝑞𝑢𝑖𝑣𝑎𝑙𝑒𝑛𝑡 𝑁𝑜𝑖𝑠𝑒 𝐵𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ

©Jack D. Peters, 27 April 2020 15 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Terminology used in “Calculating the RMS Energy of the FFT”:
Name Abbreviation Description
A numerical value that describes the Equivalent
Window Factor WF
Noise Bandwidth.

Equivalent Noise
ENBW A numerical value that describes the Window Factor.
Bandwidth

Lowest Resolvable 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛

LRF
Frequency 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛
Bin Width Bin
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛
Bandwidth BW 𝑊𝑖𝑛𝑑𝑜𝑤 𝐹𝑎𝑐𝑡𝑜𝑟
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑆𝑝𝑎𝑛
Resolution Res 2 𝑊𝑖𝑛𝑑𝑜𝑤 𝐹𝑎𝑐𝑡𝑜𝑟
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐿𝑖𝑛𝑒𝑠

Line N A single frequency measure of the Bin Width.

A general term used to describe the amplitude value

Amplitude A
of a frequency or line measured in g’s, in/s or mm/s.

A display of amplitude vs. time that represents the

Time Waveform TW
measured vibration of the machine.

An algorithm that displays amplitude vs. frequency

Fast Fourier
FFT based on a calculation that was applied to the time
Transform
waveform.

A special filter applied to the time waveform to make

Window Filter Window it periodic in the time block to simulate continuous
data when processing the FFT.

RMS Energy 𝐴 + 𝐴 + 𝐴 + ⋯𝐴
RMS
(with RMS units) 𝐸𝑞𝑢𝑖𝑣𝑎𝑙𝑒𝑛𝑡 𝑁𝑜𝑖𝑠𝑒 𝐵𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ

RMS Energy 𝐴 + 𝐴 + 𝐴 + ⋯𝐴
RMS 0.707
(with Peak units) 𝐸𝑞𝑢𝑖𝑣𝑎𝑙𝑒𝑛𝑡 𝑁𝑜𝑖𝑠𝑒 𝐵𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ

©Jack D. Peters, 27 April 2020 16 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.

 Acknowledgements:
1. Nelson L. Baxter, ABN Technical Services, Inc.
2. Dr. Ronald L. Eshleman, Vibration Institute
3. Michael Grace, Siemens Energy
4. Robert Mihata, Alta Solutions

References:
1. Eshleman, Ronald L. and Peters, Jack D., Category III Machinery Vibration Analysis,
Vibration Institute 2019.
2. Harris, Frederic J., On the Use of Windows for Harmonic Analysis with the Discrete
Fourier Transform, Proceedings of the IEEE, Volume 66, Number 1, January 1978.
3. Cerna, Michael and Harvey, Audrey F., The Fundamentals of FFT-Based Signal Analysis
and Measurement, Application Note 041, National Instruments Corporation, July 2000.
4. Dynamic Signal Analyzer Applications, Application Note 243-1, Hewlett-Packard
Company, October 1983.
5. Frarey, Jack, An Examination of Signal Resolution in an FFT Analyzer, 21st Annual
Meeting, Vibration Institute, July 17, 1997.
6. The Fundamentals of Signal Analysis, Application Note 243, Hewlett-Packard Company,
1985.

©Jack D. Peters, 27 April 2020 17 Accelent Technology LLC

Vibration Institute - Second Middle East Conference 2020 - Conference Proceedings.