Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

Using Modal Analysis and ODS Correlation

to Identify Mechanical Faults in Rotating Machinery
Brian Schwarz, Shawn Richardson, Patrick McHargue, Mark Richardson
Vibrant Technology, Inc., Centennial, CO
Click here for the IMAC Power Point presentation.
ABSTRACT
Most power plants, oil refineries, and manufacturing plants worldwide have implemented machinery health monitoring programs for
assessing the health of their rotating machinery and equipment. Digital vibration signals are the primary data used to detect and
diagnose mechanical faults in operating machinery.
A common machine fault is called “soft foot”. Soft foot is a lowering of the stiffness between a machine and its foundation.
In this paper, a modal model of a rotating machine on springs and several digital signal processing methods are used to create first-
order ODS’s for several mounting stiffnesses and internal force levels. Then, these order-based ODS’s are used by a unique
database search method called FaCTs™, to estimate the mounting stiffnesses and internal force levels in a rotating machine from
TWFs derived from cellphone videos.
FaCTs™ correlates a currently acquired ODS with ODS’s of known machine faults stored in a database. FaCTs™ can be used to
identify the location and amount of any mechanical fault based on its unique ODS. And FaCTs™ becomes more accurate as more
ODS data is acquired, labeled, and archived in a machine-based database. A typical FaCTs™ bar chart is shown in Figure 1.

Figure 1. FaCTs™ Bar Chart

KEY WORDS
Time Waveform (TWF), Digital Fourier Transform (DFT), Operating Deflection Shape (ODS), Degree of Freedom (DOF), Neural
Network (NN), Fault Correlation Tools (FaCTs™), Shape Difference Indicator (SDI), Finite Element Analysis (FEA), Auto Power
Spectrum (APS), Cross Power Spectrum (XPS), ODS-FRF (APS and Phase of an XPS relative to a reference), Structural Dynamics
Modification (SDM), Multi-Input Multi-Output (MIMO) matrix model.

Page 1 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

INTRODUCTION
Artificial Intelligence (AI) has recently become popular for interpreting the meaning of a set of data using a trained neural network. A
trained NN models the human brain, which also must be trained with input data from the moment we are born and throughout our life.
Machine learning using an NN mimics the learning of a human brain. To train an NN to identify a machine fault, it must be given lots
of data sets, each data set uniquely correlated with a mechanical fault and labeled as such. Then when given newly-acquired machine
data as input, a trained NN inference engine will identify any mechanical fault if it was trained with similar data.

FaCTs™
At Vibrant Technology, we have developed an algorithm, called FaCTs™, which functions like a trained NN, but doesn’t require
training with lots of data. Given an experimentally derived ODS, FaCTs searches a database of labeled ODS’s, each ODS labeled
with a particular machine fault. Then a FaCTs bar chart of the ten closest matching ODS’s is displayed together with the mechanical
fault associated with each labeled ODS.
FaCTs™ uses a correlation coefficient between two shapes called the Shape Difference Indicator (SDI) [11], to search a database of
archived and labeled ODS’s. By comparing the SDI value of a current ODS with each archived ODS in the database, the ten highest
SDI values and their ODS labels are displayed in a FaCTs bar chart. An example bar chart was shown in Figure 1.
• FaCTs has values between 0.0 & 1.0
• FaCTs = 1.0 ➔ two ODS’s are identical
• FaCTs above 0.9 ➔ two ODS’s are similar
• FaCTs below 0.9 ➔ two ODS’s are different
In previous papers [3], [8] we presented a new method for extracting TWFs from frames of a video. This method together with
traditional digital signal processing methods, has been used to further extract order-based ODS’s [4]-[6] of an operating machine
from a video. Using ODS’s in animation, the machine’s deformation can be visualized using frames of the video at slower speeds with
higher amplitudes.
ROTATING MACHINE
In a companion paper [10], FaCTs™ was used to uniquely identify nine different unbalance cases of the rotating machine shown in
Figure 2 using order-based ODS’s extracted from cellphone videos. This machine has a variable speed motor connected to the rotor
with a rubber belt. The motor speed was adjusted so that the rotor speed was approximately 1000 RPM throughout all the cellphone
video recordings. Those videos are also used in this paper.

Figure 2. Rotating Machine Showing Unbalance Screws Added to Its Rotors

Page 2 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

FEA Free-Body Modal Model

An FEA model with free-free boundary conditions was created for the baseplate and bearing blocks of the rotating machine. The
FEA model was then solved for its first ten modes of vibration. The deflection of an FEA bending mode shape is shown in Figure 3,
together with the modal frequencies of the first ten modes of the machine.
Because the machine was modeled as a free body in space, its first six modes have zero frequency & damping, and have rigid-body
mode shapes. Three mode shapes are rigid body deflections in three translational directions and three mode shapes are three
rotational deflections about three rotational axes.

Figure 3. Bending Mode Shape of the Base Plate and Bearing Blocks
STRUCTURAL DYNAMICS MODIFICATION (SDM)
SDM [12] is a modeling algorithm also referred to as “eigenvalue modification”. SDM calculates the new mode shapes of a
mechanical structure caused by physical modifications to the structure. Modifications are modeled with industry-standard finite
elements. The inputs and outputs of SDM are depicted in the diagram in Figure 4.

Figure 4. Structural Dynamics Modification (SDM)

In this paper, the mounting stiffnesses of the rotating machine resting on a “fixed” tabletop are modeled with four FEA spring
elements, each one attached between a corner of the base plate and the “fixed” tabletop. The four springs between the base plate and
the tabletop are also shown in Figure 3.
To model the machine attached to its fixed base with different mounting stiffness, the four FEA springs, (one between each corner of
the base plate and the “fixed” tabletop), with modeled using five different mounting stiffnesses as inputs to SDM. SDM is then used
to calculate the new modes of the machine for each stiffness case.

Page 3 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

Figure 5. First-Order ODS Animated from DFTs

ODS-FRFs
Using MEscope Video ODS™, thousands of TWFs are typically extracted from a video recording of an operating machine. A
unique frequency domain function called an ODS-FRF can be calculated from each response TWF. Not only can ODS-FRFs yield
the order-based ODS’s of rotating equipment with more accuracy, but they can be differentiated from displacement to velocity units
which are commonly used to assess vibration levels in rotating equipment.
The magnitude of an ODS-FRF is the APS of a roving response DOF of a machine. The phase is the phase of the XPS between the
response DOF and a fixed reference DOF.
An ODS-FRF carries the same engineering units as the response TWF from which was calculated. A TWF extracted from a video
has units of displacement. But because it is a frequency domain function, an ODS-FRF, (and a DFT), can be uniquely differentiated
to velocity or acceleration units. But TWF windowing and spectrum averaging can be used to reduce extraneous noise from an ODS-
FRF.
MIMO (MULTI-INPUT MULTI-OUTPUT) MATRIX MODEL
All modal analysis of the Input-Output dynamics of a mechanical structure is based on the MIMO Matrix Model shown in Figure 6.
Using the MIMO Matrix Model, Output TWFs can be calculated from FRFs & Input TWFs, Input TWFs can be calculated from
FRFs & Output TWFs, and FRFs can be calculated from Input TWFs & Output TWFs.

Figure 6 MIMO Calculation of Response TWFs from Internal Force TWFs & Mode Shapes
Page 4 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

INTERNAL FORCE TWFs FROM CELLPHONE TWFs

Next, a modal model of the machine with mounting stiffnesses of 10,000 lbf/in together with response TWFs extracted from a
cellphone video of the actual operating machine were used to calculate the internal force TWFs necessary to cause the measured
responses. This process is depicted in Figure 7A.
Both the mounting stiffnesses and the internal forces of the machine are unknown. The mode shapes resulting from five different
mounting stiffnesses would yield five different internal force TWFs for a single unbalance case. The internal unbalance force
TWFs caused by mounting stiffness of 10,000 lbf/in are displayed in Figure 7B.
Figure 7B shows that the unbalance forces in the Y direction (vertical) are about 3 times higher than the forces in the X-direction
(horizontal) at points 1 & 2 on the tops of the bearing blocks.

Figure 7A MIMO Calculation of Internal Force TWFs from Response TWFs & Mode Shapes

Figure 7B. Internal Force TWFs at Bearing Block Points 1 & 2

Page 5 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

MACHINE RESPONSE TWFs FROM FRFS & INTERNAL FORCE TWFS

Using the modal model of the machine on springs with five different mounting stiffnesses and internal force TWFs with three
different force levels, machine response output TWFs were calculated using the MIMO Matrix Model depicted in Figure 6. Typical
calculated MIMO response DFTs for points 1 & 2 on the tops of the bearing blocks is shown in Figure 8.

Figure 8. Response DFTs at Bearing Block Points 1 & 2

ARCHIVAL ODS DATABASE
The MIMO Matrix Model calculation depicted in Figure 6 was used in two calculation loops to calculate the response TWFs of the
rotating machine using five machine mounting stiffnesses and three internal force levels. The response TWFs for each stiffness and
force level case were then processed to calculate ODS-FRFs for points 1 & 2 at the top of the bearing blocks for each case.
This calculation process is depicted in Figure 9.

Figure 9. Calculation of ODS’s for Mounting Stiffnesses & Force Levels

Page 6 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

The cellphone video with four unbalance screws added to the outer rotor was chosen and labeled as the Baseline. Its TWFs were
used together with a modal model for mounting stiffnesses of 10000 lbf/in to calculate internal force TWFs for the Baseline.
Then, SDM was used together with the free-free mode shapes of the machine to calculate new mode shapes of the machine for each of
the five different mounting stiffnesses. Each set of new mode shapes was used together with one of three different force levels to
calculate response TWFs using the MIMO Matrix Model depicted in Figure 6.
The Baseline ODS together with the first-order ODS’s extracted for the ODS-FRFs for all 15 combinations of five mounting
stiffness & three internal force levels are shown in the Trend Plot Figure 10. Each ODS has four DOFs, 1X, 1Y, 2X, 2Y at points 1
& 2 at the top of each bearing block.

Figure 10. Velocity Trend Plot of Sixteen Labeled Stiffness & Force Level Cases
EVENT LOG
The Baseline ODS and the ODS’s of fifteen mounting stiffness & internal force level cases are labeled as events in the archival
database. The Event Log where the ODS’s are labeled is shown in Figure 11. Each of the sixteen cases is labeled in the Description
column.

Figure 11 Event Log of Sixteen Mounting Stiffnesses & Force Levels

Page 7 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

FaCTs™ - BASELINE CASE

When the Baseline ODS is archived into the database a second time, the FaCTs bar chart, (shown in Figure 12), clearly identifies it as
the ODS for the Baseline case.
The ODS’s for the 10000 lbf/in stiffness & 75% force level case and the 12500 lbf/in & 100 force level both closely correlate with
the Baseline ODS.

Figure 12. FaCTs™ for the Four Outboard Screws ODS

FaCTs™ - OUTBOARD ROTOR UNBALANCE CASES
Figures 13 through 15 show FaCTs correlations of ODS’s for three cases of unbalance screws added to the outboard rotor of the
rotating machine.

Figure 13. FaCTs™ for Three Outboard Screws

Page 8 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

Figure 14. FaCTs™ for Two Outboard Screws

Figure 15. FaCTs™ for One Outboard Screw

All three FaCTs bar charts in Figures 13 to 15 indicate poor correlation between the three outboard unbalances and the baseline
unbalance case. Clearly, the first-order ODS at the top of the bearing blocks is different for these cases.

Page 9 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

FaCTs™ - INBOARD ROTOR UNBALANCE CASES

Figures 16 through 18 show FaCTs correlations of ODS’s for three cases of unbalance screws added to the inboard rotor of the
rotating machine.

Figure 16. FaCTs™ for Four Inboard Screws

Figure 17. FaCTs™ for Two Inboard Screws

Page 10 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

Figure 18. FaCTs™ for One Inboard Screw

Again, all three FaCTs bar charts in Figures 16 to 18 indicate poor correlation between the three inboard unbalances and the Baseline
unbalance case. Again, the first-order ODS at the top of the bearing blocks is quite different for these three cases compared with the
Baseline ODS.
CONCLUSIONS
In this paper, several signal processing methods were used to calculate the first-order ODS at the top of the bearing blocks of a
rotating machine resulting from five different mounting stiffnesses & three different internal force levels of the machine. The free-
free mode shapes of the base plate & bearing blocks of the machine were used together with the SDM method to calculate the new
mode shapes of the machine mounted on springs with different stiffnesses between the corners of its base plate and “fixed” ground
points on a tabletop. Five different stiffness values were used for the four axial springs connected between the plate corners and the
fixed tabletop.
Using the modal model for 10000 lbf/in of a mounting stiffness, MIMO Matrix processing was used to calculate the internal force
TWFs necessary to yield the response TWFs extracted from a cellphone video of the machine. The Baseline video was recorded with
four unbalance screws attached to the outboard rotor and the machine running at about 1000 RPM. The response ODS for this
case was labeled as the Baseline ODS.
Response TWFs were then calculated for 15 different cases, using all combinations of five different mounting stiffnesses & three
different internal force levels. ODS-FRFs were then calculated from each response TWF, and the first-order ODS at 1000 RPM
was labeled and saved in an archival database.
All the digital signal processing and ODS labeling was done automatically by executing a command script in MEscope.
Finally, a unique database search method called FaCTs™ was used to correlate the ODS’s derived from cellphone videos of eight
different unbalance cases of the rotating machine with the 16 labeled ODS’s in an archival database.
The FaCTs bar charts clearly identified the Baseline ODS of the machine with four unbalance screws on the outboard rotor and
10000 lbs/in of mounting stiffness at the corners of the rotating machine. The FaCTs bar charts also showed that most of the other
unbalance screw cases were not closely correlated with any of the labeled ODS’s in the database.
The methods used in this exercise showed that mode shapes and MIMO Matrix Modeling can be used to calculate and label ODS’s,
and then use the labeled and archived ODS’s to characterize machine faults. In this paper, a common machine problem called “soft
foot” was addressed.

Page 11 of 12
Presented at IMAC XLII, January 29-February 1, 2024, Orlando, FL 2/5/2024

However, just like in the training of a neural network, much more labeled ODS data for a machine must be added to the archival
database to identify and quantify a mechanical fault such as soft foot more accurately.
Nevertheless, the precise identification of the Baseline case using FaCTs™ from the first-order ODS for only 2 DOFs at the top of
each bearing block is confirming evidence that using SDM with free-free mode shapes, MIMO Matrix Modeling, and ODS-FRF
calculations are all linear repeatable calculations. This exercise demonstrates that all these calculations can be used reliably to
characterize the linear dynamic behavior of real-world rotating machines and mechanical structures.
REFERENCES
[1] - Farnebäck G. (2003) Two-Frame Motion Estimation Based on Polynomial Expansion. In: Bigun J., Gustavsson T. (eds) Image
Analysis. SCIA 2003. Lecture Notes in Computer Science, vol 2749. Springer, Berlin, Heidelberg
[2] - https://nanonets.com/blog/optical-flow/
[3] - B. Schwarz, S. Richardson, J. Tyler, R. Spears, M. Richardson, “Post-Processing ODS Data from a Vibration Video” IMAC
XXXVIII Houston, TX, February 10-13, 2020.
[4] - M.H. Richardson, “Is It a Mode Shape or an Operating Deflection Shape?” Sound and Vibration magazine, March 1997.
[5] - B. Schwarz, M.H. Richardson, “Measurements Required for Displaying Operating Deflection Shapes” Proceedings of IMAC
XXII, January 26, 2004
[6] - B. Schwarz, M.H. Richardson, “Introduction to Operating Deflection Shapes” CSI Reliability Week, Orlando, FL, October 1999
[8] - D. Ambre, B. Schwarz, S. Richardson, M. Richardson, “Using Cellphone Videos to Diagnose Machinery Faults” IMAC XLI,
February 13-16, 2023, Austin, TX.
[9] - MEscope™ is a trademark of Vibrant Technology, Inc. www.vibetech.com
[10] - B. Schwarz, S. Richardson, P. McHargue, M. Richardson “Using Cellphone Videos to Diagnose Unbalance in Machinery”
IMAC XLII, January 29 - February 1, 2024, Orlando, TX
[11] - S. Richardson, J. Tyler, P. McHargue, M. Richardson “A New Measure of Shape Difference” IMAC XXXII February 3.6, 2014
[12] – “B. Schwarz, P. McHargue, M. Richardson “Using SDM to Train Neural Networks for Solving Modal Sensitivity Problems”
14th IMAC Conference, Dearborn, MI, February 1996

Page 12 of 12