applied

sciences
Article
Automation of Experimental Modal Analysis Using
Bayesian Optimization
Johannes Ellinger * , Leopold Beck, Maximilian Benker , Roman Hartl and Michael F. Zaeh

Institute for Machine Tools and Industrial Management (iwb), TUM School of Engineering and Design,
Technical University of Munich, Boltzmannstr. 15, 85748 Garching, Germany
* Correspondence: [email protected]

Abstract: The dynamic characterization of structures by means of modal parameters offers many
valuable insights into the vibrational behavior of these structures. However, modal parameter
estimation has traditionally required expert knowledge and cumbersome manual effort such as, for
example, the selection of poles from a stabilization diagram. Automated approaches which replace
the user inputs with a set of rules depending on the input data set have been developed to address
this shortcoming. This paper presents an alternative approach based on Bayesian optimization.
This way, the possible solution space for the modal parameter estimation is kept as widely open
as possible while ensuring a high accuracy of the final modal model. The proposed approach was
validated on both a synthetic test data set and experimental modal analysis data of a machine tool.
Furthermore, it was benchmarked against a similar tool from a well-known numerical computation
software application.

Keywords: modal parameters; modal analysis; Bayesian optimization; stabilization diagram

1. Introduction
The vibrational characteristics of structures represent one of the key design targets in
mechanical, aerospace, and civil engineering, as well as in many other domains. Typically,
the dynamic behavior is described by a linear model which relates time-varying input forces
Citation: Ellinger, J.; Beck, L.; Benker, on the structure of interest to its (also time-varying) displacements [1]. Most often, modal
M.; Hartl, R.; Zaeh, M.F. Automation models which are based on the so-called modal parameters eigenfrequencies, mode shapes,
of Experimental Modal Analysis and modal damping are used due to their intuitive physical interpretability [2]. In this
Using Bayesian Optimization. Appl. context, one also speaks of poles containing the combined information of eigenfrequency
Sci. 2023, 13, 949. https://doi.org/ and modal damping. Building up and evaluating these models is, depending on the chosen
10.3390/app13020949 approach, either called computational or experimental modal analysis (EMA). Even though
Academic Editor: Luigi Portinale simulations have improved in recent years, real-world experiments are still needed in cases
in which high accuracy is essential, or in order to calibrate the computational analyses,
Received: 11 November 2022 as, for example, demonstrated by Ellinger and Zaeh [3]. EMA is typically performed
Revised: 15 December 2022
under laboratory conditions [4], which may not be sufficient for capturing the vibrational
Accepted: 27 December 2022
characteristics of a generally nonlinear system using the linear modal model. In contrast,
Published: 10 January 2023
operational modal analysis (OMA) is performed under operational conditions, an approach
which is typically more accurate but comes with additional challenges such as only partial
or missing input force measurements [4].
Copyright: © 2023 by the authors.
Given a set of data, a variety of methods exists for estimating the modal parameters
Licensee MDPI, Basel, Switzerland. of a system. In case the modes are identified one at a time rather than all at once, they are
This article is an open access article called single-degree-of-freedom methods [5]. Popular examples include the peak-picking,
distributed under the terms and the circle-fit, and the line-fit methods [5], as well as the two-stage least squares identi-
conditions of the Creative Commons fication presented by Altintas [6]. When dealing with closely spaced modes, so-called
Attribution (CC BY) license (https:// multi-degree-of-freedom (MDOF) methods, which consider multiple modes simultane-
creativecommons.org/licenses/by/ ously, typically perform better [5]. This can, for example, be done by solving a nonlin-
4.0/). ear optimization problem [5,6], the global rational fraction polynomial method [5], the

Appl. Sci. 2023, 13, 949. https://doi.org/10.3390/app13020949 https://www.mdpi.com/journal/applsci

Appl. Sci. 2023, 13, 949 2 of 16

least squares rational function (LSRF) method [7], global singular value decomposition
(SVD) [5], the least squares complex exponential (LSCE) method [8], the least squares
complex frequency-domain (LSCF) method or its polyreference version, the so called "Poly-
MAX" method [9], or various stochastic subspace identification (SSI) methods [10]. Depend-
ing on whether they operate on time series or frequency information, these algorithms can
either be clustered as time-domain (e.g., SSI or the LSCE method) or frequency-domain
(e.g., the circle-fit method, global SVD, or the LSCF method) approaches.
The algorithms themselves are beyond the scope of this paper. However, all ap-
proaches have in common that the model order must be set, which directly relates to the
question of how many modes are to be identified. Although strategies exist for determining
a suitable model order based on the measured data, it is more common to first consider-
ably overestimate the number of modes and identify modal parameters for several model
orders and, second, to select an appropriate number of modes based on the combined
information [4,5]. This approach helps to reduce biases in the identified modes [4]. How-
ever, when proceeding in this way, many spurious or mathematical modes occur which
must be separated from the true, physical modes. The central element of this strategy is the
so-called stabilization diagram, which displays the identified modes in a model order over
frequency plot [4].
In the past, selecting appropriate modes from the stabilization diagram has required
a high level of user interaction and expert knowledge [11,12]. To simplify the process,
Scionti et al. [13] first made use of several methods to eliminate spurious modes applicable
to state spaced models and, second, introduced a set of rules to mimic the selection of
modes by the user. The rules depend on several parameters, for which default values were
set. They found that the proposed approach performs well on in-flight flutter data for an
aircraft. A short review on such mode separation rules, which, in the end, emulate the
actions and decisions of an experienced user, was provided by van der Auweraer and
Peeters [14]. The latter further mentioned in-flight data analysis and structural health
monitoring as possible applications. Their work was extended by Lau et al. [1], who
also referred to Scionti et al. [13] by developing a rule-based automatic modal parameter
selection approach. Peer group comparisons between modal analysis novices and experts
on two comprehensive data sets proved that the approach leads to skill-independent results
and an overall productivity gain of 50%.
A different approach was followed by Scionti and Lanslots [11], who compared several
Fuzzy c-means clustering techniques with and without an additional genetic algorithm (GA)
for the initialization of finding pole candidates. It was found that the Gustafson–Kessel
algorithm with initial values from a GA resulted in the best clusters of poles. Some clusters
were immediately discarded based on their contribution ratio, their compactness measures,
and corresponding thresholds. For each remaining cluster, the pole closest to its center
was chosen as the final result, which was positively evaluated by comparing it to the poles
found by experts.
All of the works presented still rely on user-defined thresholds and are thus not
fully automated [12]. This drawback was overcome by Reynders et al. [12], who, in
addition to a very comprehensive literature review, proposed a three-stage approach for
fully automated modal parameter estimation. First, all modes were classified as either
spurious or potentially stable by eleven soft and three hard validation criteria. Second,
the remaining poles were hierarchically clustered to find similar modes and, finally, the
mode with a modal damping closest to the median damping of the corresponding cluster
was selected. The approach was successfully tested on several data sets. However, the
results were only benchmarked against modal parameters retrieved by a manual analysis. A
comparison of the modal model with the original data set was not performed. A very similar
strategy was followed by Neu et al. [15]. However, the latter allowed larger damping values,
used different features representing the soft validation criteria, and selected the final poles
by averaging all pole candidates per cluster. Additionally, they examined the influence
of the maximum model order and found that their approach was less sensitive to it than
Appl. Sci. 2023, 13, 949 3 of 16

that by Reynders et al. [12]. The method was applied to an OMA data set and successfully
compared to EMA and finite element analysis results. Mugnaini et al. [16] proposed slight
adaptations to the approach of Neu et al. [15] and proved their effectiveness in an ablation
study for a helicopter blade data set. Additionally, the eigenfrequency identification was
validated on a simulated data set. Again, no validation against the original data set was
conducted in Neu et al. [15] and Mugnaini et al. [16].
It can be concluded that a consensus has been formed in the state-of-the-art regarding
the basic approach of automatically extracting modal parameters [12,15,16]. However, it
was not evaluated how well the identified modal model replicates the original data set
using, for example, the frequency response assurance criterion (FRAC), nor was this the
focus of the modal parameter extraction. Furthermore, the parameters which required
user input in earlier approaches [1,11,13,14] were replaced by metrics performed on the
input data set [12,15,16], which is believed to unnecessarily restrict the solution space for
the final modal parameters. The present paper presents an alternative approach, in which
Bayesian optimization is used to find the optimal values of any required hyperparameters.
The deviation between the identified modal model and the original data set is used as a
cost function. This also ensures a high level of validity for the final modal model.
The remainder of this paper is structured as follows: Section 2 presents the proposed
approach in general. It will be shown how modal parameters are extracted based on a few
hyperparameters, which will subsequently be optimized using a Bayesian optimization
approach. Section 3, on the one hand, validates the proposed approach on a synthetic test
data set. On the other hand, modal parameters were extracted from input data measured
on a machine tool. Additionally, the results from both data sets were benchmarked against
the MATLAB® function modalfit. Last, Section 4 summarizes the main content of the
paper and offers an outlook to future research.

2. Proposed Method
This section presents a new approach for modal parameter estimation, which will be
called AutoEMA. In contrast to the recent literature [12,15,16], hyperparameters are not
automatically derived from the input data set, but rather set by a Bayesian optimization
algorithm such that the match between the identified modal model and the original input
data is as high as possible. Section 2.1 describes in detail how modal parameters are
extracted based on (measured) input data and a set of hyperparameters. Section 2.2
demonstrates how values for these hyperparameters can be found such that the modal
parameters and the corresponding modal model optimally represent the original input
data. Figure 1 illustrates the overall approach.

2.1. Modal Parameter Estimation

The first step is calculating pole candidates for all model orders pi = 1, 2, . . . , pmax
up to and including the maximum model order pmax . In the present research, this is
done using the PolyMAX algorithm [9], but a variety of other methods are also applicable
(see Section 1). The found pole candidates are then classified as certainly spurious and
potentially physical by applying four criteria: The relative difference in frequency of a
pole of order pi to any pole of model order pi−1 must be smaller than k ∆ f (a), the relative
difference in damping of a pole of order pi to any pole of model order pi−1 must be smaller
than k ∆D (b), and the pole’s damping must be larger than k Dmin (c) and smaller than k Dmax
(d). If all conditions (a) to (d) for a given pole are met, it is classified as potentially physical,
otherwise as certainly spurious. Figure 2 exemplarily shows the frequency of all potentially
physical poles of a given frequency response function (FRF).
Appl.
Appl. Sci. 13,949
2023,13,
Sci.2023, 949 44of
of16
16

Yes
Calculation Modal No
Start of new hyper- parameter i ≤ nmax ? End
parameters estimation

given the hyperparameters pmax , k ∆ f , k ∆D , k Dmin , k Dmax , dmax , qmin , and k eukl

Classification Pole selection

Pole Clustering
Start (into physical (and mode shape End
calculation
and spurious) calculation)

Figure 1.
Figure Flowchart showing the AutoEMA method; the modal parameter
1. Flowchart parameter estimation
estimation is
is based
based on
on aa
set of hyperparameters
set of hyperparameters explained in Table 1. Their value is determined by a Bayesian optimization
determined by a Bayesian optimization
strategy such
strategy such that
that the
the resulting
resulting model
model optimally
optimally replicates
replicates the
the original
original input
input data.
data.

Reference
Reference 140
140
→

Fitted
m/N→

10−
10
−77 Fitted
120
120
inm/N

Potentially
Potentially physical
physical pole
pole
100
100

→
amplitudein

order→
Receptanceamplitude

Modelorder
80
80

Model
60
60
−88
10−
Receptance

10
40
40

20
20

00
00 100
100 200
200 300
300 400
400 500
500
Frequency in Hz
Frequency in Hz →
→

Figure 2.
Figure Exemplarily FRF
2. Exemplarily FRF and
and potentially
potentially physical
physical poles
poles identified
identified after
after applying
applying the
the PolyMAX
PolyMAX
algorithm and criteria (a) to (d) (see Section 2.1) for the given hyperparameters k∆∆ff,, kk∆D
algorithm and criteria (a) to (d) (see Section 2.1) for the given hyperparameters k , k Dmin ,
∆D , k Dmin ,
and kk Dmax..
and Dmax

Spurious poles
Spurious poles are
are discarded
discarded and and thus
thus excluded
excluded fromfrom further
further analysis.
analysis. Given
Given that
that
many model orders were considered, there are generally several remaining poles for
many model orders were considered, there are generally several remaining poles for any
any
mode shape
mode shape being
being identified.
identified. Agglomerative
Agglomerative clustering
clustering is is used
used to
to group
group poles
poles into
into modes.
modes.
Given that
Given that this
this is
is aa hierarchical
hierarchical clustering
clustering algorithm,
algorithm, the the algorithm
algorithm determines
determines the
the number
number
of clusters.
of clusters. However,
However, aa threshold
threshold has has to
to be
be set,
set, in
in this
this case
case the
the maximum
maximum frequency
frequency distance
distance
of poles
of poles within
within aa cluster
cluster ddmax .. If
If the
the clustering
clustering isis applied
applied toto the
the poles
poles in
in Figure
Figure 2,
2, six
six possible
possible
max
modes are found, as illustrated in
modes are found, as illustrated in Figure 3. Figure 3.
Two more
Two more rules
rules were
were applied
applied in in order
order to
to avoid
avoid clusters
clusters having
having spurious
spurious poles
poles only.
only.
First, it was assumed that clusters reflecting actual modes contain at least a certain percent-
First, it was assumed that clusters reflecting actual modes contain at least a certain percent-
age of the model order p of poles, in this case q . This rule directly corresponds to the
age of the model order pii of poles, in this case qmin min . This rule directly corresponds to the
original assumption that physical poles show up at many model orders [14]. Second, the
original assumption that physical poles show up at many model orders [14]. Second, the
variance in frequency of poles of spurious-only clusters is assumed to be higher than the
variance in frequency of poles of spurious-only clusters is assumed to be higher than the
one from a cluster reflecting an actual mode. Thus, all clusters with an average Euclidean
one from a cluster reflecting an actual mode. Thus, all clusters with an average Euclidean
distance larger than k eukl · dmax are discarded, where dmax is the maximum frequency dis-
distance larger than k eukl · dmax are discarded, where dmax is the maximum frequency dis-
tance of poles within a cluster from the clustering stage and k is an euclidean frequency
tance of poles within a cluster from the clustering stage and keukl eukl is an euclidean frequency
distance factor.
distance factor.
Appl. Sci. 2023, 13, 949 5 of 16

14
Identified modes
Selected poles

Modal damping ratio in % →

12

10

2
50 100 150 200 250 300 350
Frequency in Hz →

Figure 3. Identified modes by agglomerative clustering with a maximum frequency distance dmax of
poles within a cluster; different clusters, that is, distinct modes, are indicated by different colors. The
finally selected poles are marked with a cross.

Finally, one pole per cluster can be selected by choosing the pole closest to the median
pole location of a cluster, which is indicated by crosses in Figure 3. The pole determines
both eigenfrequency and modal damping of the mode corresponding to the cluster. After all
poles have been selected, the mode shapes are calculated using the least squares frequency-
domain method [9].
Table 1 summarizes the hyperparameters involved in the modal parameter estimation.
The code used in this work is based on the Python package pyEMA [17].

Table 1. Overview of hyperparameters used in the modal parameter estimation; the classification
parameters are defined similar to Reynders et al. [12] and Neu et al. [15].

Step Symbol Description

Pole calculation pmax Maximum model order
Classification k∆ f Maximum relative frequency difference of two poles
k ∆D Maximum relative damping difference of two poles
k Dmin Minimum damping of a mode
k Dmax Maximum damping of a mode
Clustering dmax Maximum frequency distance within a cluster
qmin Minimum percentage of the model order of poles per cluster
k eukl Maximum euclidean frequency distance factor between poles
Pole selection - -

2.2. Hyperparameter Optimization

The modes identified and the corresponding modal parameters strongly depend on
the hyperparameters used in the modal parameter estimation step (see Table 1). The key
idea behind this publication is to find optimal values for these parameters, whereas the
recent literature has dealt with deriving them based on the input data set [12,15,16]. On
the one hand, the presented approach keeps the solution space as widely open as possible.
On the other hand, it ensures a high concordance between the original input data and
the resulting modal model. A variety of optimization algorithms exist in the literature.
The decision on which depends, amongst others, on the amount of knowledge on the
system being optimized, the availability of efficient gradient calculations, and the shape as
well as the (computation) costs of evaluating the objective function. The specific objective
Appl. Sci. 2023, 13, 949 6 of 16

function used will be described below and will involve running the modal parameter
estimation (see Section 2.1). As the relationship between the hyperparameters and the
resulting modal parameters is rather complex, no gradients are available since pmax is a
discrete number, the objective function is generally non-convex, and the costs of evaluating
the objective function can be high (depending on the chosen maximum model order pmax
and the size of the input FRF data set), the Bayesian optimization approach was chosen
here [18]. In the context of this method, the objective function is generally treated as random
and approximated with a Gaussian process regression model. Previous objective function
evaluations are treated as prior knowledge and are considered as a prior. New function
evaluations increase the knowledge about the objective function, which is then used to
update the prior, forming the so-called posterior. The posterior, in turn, can then be used
in several ways to determine the location of further new, promising function evaluations.
Those locations are determined with the help of an acquisition function. In this case, the
so-called upper confidence bound optimization policy was applied, allowing the balance
between exploitation and exploration to be easily set. A comprehensive guide on Bayesian
optimization can be found in Garnett [18].
The AutoEMA method tries to find values for the hyperparameters (see Table 1) such
that the identified modal parameters and the corresponding modal model best replicate
the original input data, which is given in the form of FRFs. To measure the match between
two FRFs, the FRAC can be used:

| H1 (ω )T H2∗ (ω )|2
FRAC( H1 (ω ), H2 (ω )) =
, (1)
H1 (ω ) T H1∗ (ω ) H2 (ω ) T H2∗ (ω )

where H1 (ω ) and H2 (ω ) are the FRFs being compared, ω is the angular frequency, and x∗
is the complex conjugation and x T the transpose of a vector x [19]. FRAC values of 100%
indicate a perfect match, whereas values of 0% mean no correlation at all.
An average value over all input FRFs is computed to assess the complete input data set:
ninp
1
FRAC =
ninp ∑ FRAC( Hinp,k , Hmodel,k ) (2)
k =1

In this case, ninp is the number of input FRFs, Hinp,k are the original input FRFs, and
Hmodel,k are the FRFs resulting from the modal model. Note that the angular frequency ω
has been dropped for the sake of readability. The concordance between the model FRFs
and the input FRFs generally increases with an increasing number of modes in the model.
To prevent spurious modes approximating noise in the measurement data, Equation (2) is
extended by a (small) regularization parameter r, which penalizes a high number of found
modes nmodes :

y = FRAC − nmodes r (3)

The final optimization cost function y can then be constructed by combining

Equations (1)–(3), with Hmodel,k in Equation (2) being a function of the hyperparameters to
be optimized:

Hmodel,k = f (k, pmax , k ∆ f , k ∆D , k Dmin , k Dmax , dmax , qmin , k eukl ) (4)

As shown in Figure 1, the Bayesian optimization is terminated when the current

iteration number i exceeds the set maximum iteration number nmax .

3. Application
In this section, the application of the approach presented in Section 2 is described for
two use cases: First, synthetic test data were simulated using a sixth-order MDOF oscillator
model (see Section 3.1). In this case, the ground truth values of the modal parameters
Appl. Sci. 2023, 13, 949 7 of 16

were known per definition, and their estimated values could be directly compared to
their reference values. Artificial noise was superimposed to show the robustness of the
proposed approach. Second, data measured on a real-world machine tool test bench were
used as input for the automated modal parameter estimation approach (see Section 3.2).
In both cases, the resulting modal parameters were benchmarked against the function
modalfit from the signal processing toolbox of the well-established numerical computation
program MATLAB® .

3.1. Synthetic Test Data

To test the proposed approach, an oscillator with six degrees of freedom (DOFs)
was set up as shown in Figure 4 and used for simulating synthetic reference input data.
Using a time-domain simulation, six FRFs were calculated from an input force at the first
mass f 1 to the displacement at all DOFs. To demonstrate the robustness of the AutoEMA
method, normal (Gaussian) noise with zero mean was superimposed to the respective
outputs x1 , x2 , . . . , x6 . Additionally, the (damped) eigenvalue problem was solved, resulting
in estimates for the mode shapes, the eigenfrequencies, and the modal damping ratios.
These modal parameters are, in the following, regarded as ground truth values and to be
estimated using both the MATLAB® and the AutoEMA approach.

x1 , f 1 x2 , f 2 xn , f n

c1 c2 cn
···
m1 m2 mn
···
d1 d2 dn

Figure 4. Schematic illustration of an MDOF system used for generating synthetic measurement data
in Section 3.1.

Regarding the AutoEMA approach, Table 2 shows the value range constraining the
hyperparameters of the modal parameter estimation in the Bayesian optimization process
and their initial values. These boundaries were considered to be wide in order to keep the
solution space as widely open as possible. Note that the minimum damping of a mode
k Dmin was set to the lowest physically meaningful value of 0%, so it was excluded from the
optimization in this case. Per definition, the regularization parameter r cannot be optimized
and was also fixed. However, the value r of 0.002 has been found to be suitable for many
data sets, including the one used herein and the real-world data shown in Section 3.2.
The initial values were chosen arbitrarily within the bounds, as the result of the Bayesian
optimization was found to be widely insensitive. In total, 40 iterations, that is, evaluations
of the cost function (see Section 2.2), were made.
The same data were also fed into the MATLAB® function modalfit. It requires the
number of modes to be determined as an input parameter. In this case, it was set to the true
value of six modes, which was also identified automatically by the AutoEMA approach.
The modalfit function has implemented both the LSCE and the LSRF algorithms. Here,
the default LSCE method was used.
Appl. Sci. 2023, 13, 949 8 of 16

Table 2. Value range and initial values of the hyperparameters used in the optimization stage of the
modal parameter optimization for both use cases; a description of the parameters can be found in
Table 1 and Section 2.1. The regularization parameter r and the minimum modal damping k Dmin
were set constant.

Hyperparameter Initial Value Value Range

pmax 80 60 to 120
k∆ f 1% 0.1% to 10%
k ∆D 5% 5% to 20%
k Dmin 0%
k Dmax 20% 20% to 30%
dmax 2 Hz 0.4 Hz to 4 Hz
qmin 30% 20% to 60%
k eukl 0.5 0.1 to 0.8
r 0.002

Figure 5 shows the (simulated) reference FRF from a force input f 1 at the first DOF
to the displacement x1 at the same DOF alongside the reconstructed FRFs from both the
MATLAB® and AutoEMA modal models. It can be seen that both approaches estimated
the original FRF well despite the presence of noise. However, the AutoEMA approach
slightly outperformed the MATLAB® approach in the low frequency region up to 150 Hz.
Receptance amplitude

10−7
in m/N →

10−8

10−9

Reference
Phase in deg →

100 Matlab
AutoEMA
0

−100

0 50 100 150 200 250 300 350

Frequency in Hz →

Figure 5. Comparison of the receptance amplitude and the phase between the input (solid) and the
fitted FRF (AutoEMA: dashed, MATLAB® : dash-dotted) from the modal model from a force f 1 to the
displacement x1 of the first mass; normal (Gaussian) noise was added to the (synthetic) displacement
data to simulate noisy real-world measurements.

Figure 6 shows the conformance of the estimated mode shapes with the (simulated)
reference data quantified by the modal assurance criterion (MAC) [20], which is defined as

|ϕT1 ϕ∗2 |2
MAC(ϕ1 , ϕ2 ) =
. (5)
ϕT1 ϕ∗1 ϕT2 ϕ∗2

In this case, ϕi are mode shapes being compared, and, again, x∗ denotes the complex
conjugation and x T the transpose of a vector x. Similar to the FRAC, a value of 0% indicates
no correlation at all, and a result of 100% states a perfect coincidence between the two mode
Appl. Sci. 2023, 13, 949 9 of 16

shapes. It can be seen that both approaches estimated the reference mode shapes equally
well with very high concordance.

Mode number (Reference) →

6 97.9

Mode number (Reference) →

6 99.2
80 80

MAC value in % →
MAC value in % →
5 97.5 5 99.8

4 99.9 60 4 99.9 60

3 100.0 40 3 100.0 40

2 100.0 2 100.0
20 20
1 100.0 1 100.0

1 2 3 4 5 6 1 2 3 4 5 6
Mode number (MATLAB ® ) → Mode number (AutoEMA) →

(a) (b)

Figure 6. Comparison of the found mode shapes using the MAC; here, the mode shapes from solving
the damped eigenvalue problem resulting from the MDOF oscillator in Section 3.1 were used as
ground truth, (a) MATLAB® modalfit, (b) AutoEMA.

To assess the accuracy of eigenfrequencies f ie , the natural frequency difference (NFD)

criterion can be used [21]:

| f 1e − f 2e |
NFD( f 1e , f 2e ) = (6)
min f 1e , f 2e

By slightly adapting Equation (6), a similar comparison can be made regarding the
modal damping ratios ξ i using the natural damping difference (NDD):

|ξ 1 − ξ 2 |
NDD(ξ 1 , ξ 2 ) = (7)
min(ξ 1 , ξ 2 )

Both criteria indicate perfect coincidence with a value of 0% and increase with a rising
difference between the damping ratios and eigenfrequencies to be compared, respectively.
Tables 3 and 4 show these criteria along with the absolute values for the synthetic data use
case. Again, it can be seen that both approaches estimated the reference modal damping
ratios and eigenfrequencies well with maximum deviations of 0.3% and 17.1% of the NFD
and NDD, respectively.

Table 3. Overview of the eigenfrequency results for the synthetic data use case (see Section 3.1)
showing the absolute values and the NFD for both the MATLAB® (ML) and AutoEMA (AE) approach.

Mode ωre f ωM L ω AE NFDre f − M L NFDre f − AE

1 33.8 Hz 33.8 Hz 33.7 Hz 0.0% 0.2%
2 86.6 Hz 86.6 Hz 86.6 Hz 0.0% 0.0%
3 186.7 Hz 186.6 Hz 186.6 Hz 0.0% 0.0%
4 237.3 Hz 236.6 Hz 237.3 Hz 0.3% 0.0%
5 250.5 Hz 250.3 Hz 250.5 Hz 0.1% 0.0%
6 318.9 Hz 318.4 Hz 319.5 Hz 0.1% 0.2%
Appl. Sci. 2023, 13, 949 10 of 16

Table 4. Overview of the modal damping results for the synthetic data use case (see Section 3.1) show-
ing the absolute values and the NDD for both the MATLAB® (ML) and AutoEMA (AE) approach.

Mode ξ re f ξ ML ξ AE NDDre f − M L NDDre f − AE

1 11.4% 11.3% 10% 0.7% 13.9%
2 5.0% 5.0% 4.8% 0.3% 3.9%
3 3.4% 3.4% 3.4% 0.2% 0.1%
4 3.3% 3.6% 3.3% 8.6% 0.6%
5 3.3% 3.4% 3.2% 4.4% 1.9%
6 3.5% 3.7% 3.0% 6.1% 17.1%

3.2. Machine Tool Data

A second data set was acquired on a DMG DMC duo Block 55H machine tool in
order to demonstrate the robustness and effectiveness of the proposed modal parameter
extraction method (i.e., the AutoEMA method). As illustrated in Figure 7, the machine tool
consists of a machine tool bed and a workpiece table (WPT), enabling movement in the
global z-axis.
N27
y N19

z
x

N5
N18

N1 , N11
N0 , N9

N28 N17

N2 , N10
N3 , N8
z4

z3

z2

z1

N22

N20 N21

Machine bed

Workpiece table (WPT)

Figure 7. Rendering of the considered machine tool, the measurement positions z1 to z4 , and the
measured nodes; the WPT is depicted in position z3 and a description of the node points can be found
in Table 5.

The machine tool’s vibrational response was measured at 17 nodes in all three spatial
directions. Four Kistler® triaxial accelerometers (two times type 8762A10 and two times
type 8762A50) were used together with a National Instruments (NI)® cDAQ-9198 rack
with three type NI® -9232 modules and one type NI® -9234 module. The measurements
were repeated for four WPT positions z1 , z2 , z3 , and z4 along the z-axis with impulse
hammer excitations in x-, y-, and z-directions. The node points are shown in Figure 7 and
described in Table 5. In particular, a corner of the machine tool bed (N28 ) was chosen as
the excitation node. For each WPT position, the data acquisition has resulted in 153 FRFs
Appl. Sci. 2023, 13, 949 11 of 16

(17 nodes measured in three spatial directions for three excitation directions). The FRFs
were calculated as the average of three hammer hits for a measurement duration of 4 s with
a sampling rate of 10.240 Hz.

Table 5. Model nodes considered; the node locations are illustrated in Figure 7.

Node Description
N0 , N9 Shoe and rail nodes of the first LGS shoe
N1 , N11 Shoe and rail nodes of the second LGS shoe
N2 , N10 Shoe and rail nodes of the third LGS shoe
N3 , N8 Shoe and rail nodes of the fourth LGS shoe
N5 WPT node
N17 , N18 , N19 , N27 Machine bed nodes
N20 , N21 , N22 ME nodes
N28 Excitation node

The approach presented in Section 2 was applied for each WPT position using the
same parameter bounds as in the synthetic data use case (see Table 2). Because of the
more complicated structure, the optimization was stopped after 100 iterations in this case.
Table 6 shows the final values of the hyperparameters (see Table 1) for all considered
WPT positions. It can be seen that most parameters have been estimated quite similarly
over all WPT positions. However, some outliers exist as, for example, the maximum
model order pmax for WPT position z4 . The data set contained redundant measurements
since, for each WPT position, three measurements for each DOF resulting from three
different excitation directions had been made. It is believed that redundancy has supported
the estimation of the eigenfrequencies and the modal damping ratios. However, the
mode shape information was only extracted from measurements with an excitation in the
y-direction as the input force was the easiest to control in that direction, thus leading to the
best mode shape estimations.

Table 6. Found values of the hyperparameters (see Table 1) in the Bayesian optimization step for
different WPT positions.

pmax k∆ f k∆D k Dmax dmax qmin keukl

z1 108 4.2% 15.1% 25.6% 0.20 Hz 27.9% 0.50
z2 108 5.9% 6.9% 21.7% 0.52 Hz 28.8% 0.61
z3 108 0.4% 6.2% 24.0% 0.36 Hz 28.6% 0.55
z4 112 5.4% 11.0% 24.2% 0.58 Hz 28.2% 1.64

The same data were also fed into the MATLAB® modalfit function, which requires the
number of modes being determined as an input parameter. In order to obtain comparable
results, it was set to the same number which the AutoEMA approach has found. For
this use case, no meaningful results could be produced with the LSCE algorithm, leaving
only the LSRF method for the evaluations. It is noteworthy that the modal parameter
estimation using the AutoEMA method took less than one minute on a laptop with four
Intel® i7-7700HQ CPU cores, whereas a run of the LSRF algorithm required 22 h and 48 min
on a simulation workstation with 24 Intel® Xeon® Gold 5220R CPU cores. As a result, the
MATLAB® approach was only run for WPT position z2 .
Figure 8 shows the correlation between the mode shapes found by MATLAB® and
the proposed approach for the frequency range up to 300 Hz, which was determined by
the MAC. It can be seen that most modes were found by both methods, indicated by MAC
values higher than 80%. However, some modes were only found by the AutoEMA method
but not by the MATLAB® function, and vice versa. Given that, as will be shown below,
the match between the modal models and the input data was very high in both cases, this
result indicates modes that were either not well observed or highly damped.
Appl. Sci. 2023, 13, 949 12 of 16

17 (263.1 Hz)
16 (251.6 Hz)
15 (238.9 Hz) 90.6
14 (225.9 Hz) 98.5 80
13 (207.1 Hz)
Mode number (AutoEMA) →
12 (187.7 Hz)

MAC value in % →
11 (180.2 Hz) 97.5 25.6
60
10 (160.7 Hz)
09 (145.1 Hz)
08 (120.3 Hz) 97.5
40
07 (108.0 Hz) 99.9
06 (095.0 Hz) 98.0
05 (075.7 Hz) 94.5
04 (052.0 Hz) 25.397.2 20
03 (041.9 Hz) 96.237.7
02 (031.0 Hz) 99.796.8
01 (028.7 Hz) 46.650.5 10.8
01 (028.8 Hz)
02 (029.1 Hz)
03 (031.3 Hz)
04 (031.5 Hz)
05 (042.3 Hz)
06 (043.1 Hz)
07 (049.3 Hz)
08 (052.3 Hz)
09 (053.4 Hz)
10 (076.1 Hz)
11 (095.5 Hz)
12 (108.2 Hz)
13 (120.1 Hz)
14 (180.3 Hz)
15 (226.7 Hz)
16 (240.2 Hz)
17 (255.5 Hz)
Mode number (MATLAB ® ) →
Figure 8. Correlation between the modes up to 300 Hz found by the AutoEMA approach and
MATLAB® modalfit determined by the MAC for the WPT position z2 .

Figure 9 qualitatively illustrates the resulting match between the mean input FRFs and
the reconstructed mean FRFs from the modal model for two WPT positions. It can be seen
that there is a very high level of concordance between the input and fitted data for both
positions for the AutoEMA approach, thus highlighting the validity of the model. However,
there were higher deviations for the MATLAB® approach at position z2 , especially in the
high frequency region. This outcome was also confirmed by the mean FRAC values of
88.7% for the AutoEMA FRF at position z2 and 64.8% for the MATLAB® FRF, respectively.
In general, even FRAC values of 70% are considered to be a good match [22].
The same conclusion can be drawn by looking at the individual FRFs in Figure 10.
It can be seen that, for both approaches, the direct driving point FRF from a force at node N28
in the y-direction to the measured displacement at the same node in the same coordinate
direction and the FRF from a force at node N28 in the y-direction to the measured deflection
at one of the edges of the machine tool (node N19 ) in z-direction were approximated very
well by the modal models. Regarding the AutoEMA approach, the same was also true
for other FRFs, which is shown in Table 7: Only for 10% of the measured FRFs at all WPT
positions, the modal model leads to a fitted FRF with, measured by the FRAC value, less
than 82.6% concordance. However, the MATLAB® approach on average approximates
the input FRFs only with a match of 64.8% and one-tenth of them even worse than with
18.9% concordance.
Appl. Sci. 2023, 13, 949 13 of 16

Receptance amplitude
Position z2
Position z4
10−8

in m/N →
10−9

50 100 150 200 250 300

Frequency in Hz →

Figure 9. Comparison of the mean receptance amplitude of the original input FRFs (solid) with the
fitted FRFs from the modal model (MATLAB® : dash-dotted, AutoEMA: dashed); the comparison
is shown for the WPT positions having the lowest number of found modes z2 and the one at the
greatest distance, that is, position z4 . Due to the high calculation times, MATLAB® results are only
available for position z2 .
Receptance amplitude

10−7
in m/N →

10−9

(y) (y)
N28 → N28
Phase in deg →

100
(y) (z)
N28 → N19
0

−100

50 100 150 200 250 300

Frequency in Hz →

Figure 10. Comparison of the receptance amplitude and the phase between measured (solid) and
fitted FRFs (MATLAB® : dash-dotted, AutoEMA: dashed) from the modal model for position z2 ;
(β ) (β )
Nα1 1 → Nα2 2 denotes the FRF between a force excitation at node Nα1 in coordinate direction β 1 and
the measured displacement at node Nα2 in coordinate direction β 2 (see Table 5).

Table 7. FRAC value statistics for FRFs measured for WPT position z2 for the machine tool use case.

10%
Minimum Mean Median Maximum
Percentile
AutoEMA 61.0% 82.0% 88.6% 90.1% 94.7%
MATLAB ® 0.1% 18.9% 64.8% 73.3% 99.7%

Table 8 shows a comparison of eigenfrequencies estimated by the MATLAB® and the

AutoEMA approach for the machine tool data use case. It can be seen that both approaches
estimate very similar eigenfrequencies with a maximum NFD (see Equation (6)) of only
1.2%. This holds especially true considering that even a mean relative eigenfrequency dif-
ference of 3.3% is sufficient for many applications [23]. However, the MATLAB® approach
seemed to generally estimate higher modal damping ratios with NDDs up to 213.7%, as
can be seen in Table 9.
Appl. Sci. 2023, 13, 949 14 of 16

Table 8. Comparison of eigenfrequencies f ie found by MATLAB ® and AutoEMA for WPT position
z2 ; "-" indicates that no matching mode with a MAC value of at least 80% was found.

f1e f2e f3e f4e f5e f6e f7e f8e

AutoEMA 28.8 Hz 31.0 Hz 41.8 Hz 52.1 Hz 75.7 Hz 95.0 Hz 108.1 Hz 120.0 Hz
MATLAB ® - 31.3 Hz 42.3 Hz 52.3 Hz 76.1 Hz 95.5 Hz 108.2 Hz 120.1 Hz
NFD AE− ML - 1.1% 1.2% 0.3% 0.4% 0.6% 0.1% 0.1%

Table 9. Comparison of modal damping ratios ξ i found by MATLAB ® and AutoEMA for WPT
position z2 ; "-" indicates that no matching mode with a MAC value of at least 80% was found.

ξ1 ξ2 ξ3 ξ4 ξ5 ξ6 ξ7 ξ8
AutoEMA 1.05% 0.88% 1.67% 1.04% 1.98% 3.82% 0.29% 1.50%
MATLAB ® - 2.78% 2.75% 1.68% 2.22% 3.20% 0.28% 3.02%
NDD AE− ML - 213.7% 64.8% 62.2% 12.2% 19.2% 5.8% 101.0%

Tables 10 and 11 show the first eight identified eigenfrequencies and modal damping
ratios using the AutoEMA method for WPT position z2 , which is the one with the lowest
number of identified eigenmodes (26 modes up to 300Hz). Similar modes at the other
positions were found using the MAC [24], and their eigenfrequencies and damping ratios
are also listed. It can be seen that the modes 1 to 4 exhibited only a low level of dependency
on the WPT position, whereas the eigenfrequency and modal damping ratios of modes 5 to
8 changed from position z1 to position z4 . This result can also be seen in the comparison of
the mean measured and fitted FRFs in Figure 9.

Table 10. Identified first eight eigenfrequencies f ie for the WPT position z2 with the lowest number of
found modes and the equivalent modes’ eigenfrequencies at the other positions; “-” indicates that no
matching mode with a MAC value of at least 80% was found.

f1e f2e f3e f4e f5e f6e f7e f8e

z1 28.7 Hz 31.0 Hz - 52.0 Hz 76.1 Hz 91.9 Hz 109.1 Hz 117.3 Hz
z2 28.8 Hz 31.0 Hz 41.8 Hz 52.1 Hz 75.7 Hz 95.0 Hz 108.1 Hz 120.0 Hz
z3 28.7 Hz 31.0 Hz 41.4 Hz 52.1 Hz 75.2 Hz 97.9 Hz 107.0 Hz 122.4 Hz
z4 - 31.0 Hz - 52.1 Hz 75.0 Hz - 106.1 Hz 127.3 Hz

Table 11. Identified first eight modal damping ratios ξ i for the WPT position z2 with the lowest
number of found modes and the equivalent modes’ damping ratios at the other positions; “-” indicates
that no matching mode with a MAC value of at least 80% was found.

ξ1 ξ2 ξ3 ξ4 ξ5 ξ6 ξ7 ξ8
z1 1.04% 0.85% - 1.18% 1.93% 3.59% 0.33% 1.54%
z2 1.05% 0.88% 1.67% 1.04% 1.98% 3.82% 0.29% 1.50%
z3 1.12% 0.86% 2.34% 1.03% 1.98% 2.27% 0.29% 2.66%
z4 - 0.96% - 1.01% 2.16% - 0.32% 2.93%

4. Conclusions and Outlook

This paper presented an approach for automatically estimating the modal parameters
of a system based on input measurements. In contrast to the state-of-the-art, no user input
is needed for evaluating the stabilization diagram, nor is the solution space restricted
by fixed rules derived from the input data set. Instead, a Bayesian optimization process
ensures that modal parameters are found, thus leading to the model best representing the
original input data. The proposed AutoEMA approach was first theoretically derived in
Section 2, and was validated using simulated input data from an MDOF oscillator. The
approach was benchmarked against ground truth reference values and the MATLAB®
modalfit function in this context. Finally, the proposed approach was also applied to EMA
data from a machine tool test bench. It was found that the AutoEMA approach led to
modal parameters which effectively represent the original input data for different machine
Appl. Sci. 2023, 13, 949 15 of 16

tool axis positions, thereby outperforming the MATLAB® method in both accuracy and
calculation time.
Although this approach has already been successfully tested on two very different
data sets (see Section 3), future research will further validate the proposed approach. Data
from domains other than machine tools will be used for this purpose. Special attention will
be paid to very closely spaced modes, symmetrical modes, and heavily coupled modes.
Currently, the presented approach is limited to EMA with FRFs as input data. It will be
examined in the future if and how it can be extended to work with OMA and, for example,
power spectral densities as input data. Moreover, the AutoEMA method now offers a
means of reliably extracting modal parameters in a computationally efficient manner. It
will be researched how this information can be used best in parameter identification of
machine tool models.

Author Contributions: Conceptualization, J.E., L.B. and M.B.; methodology, J.E., L.B. and M.B.;
software, L.B., J.E. and M.B.; validation, J.E., M.B. and L.B.; investigation, L.B., J.E. and M.B.; resources,
J.E., R.H. and M.F.Z.; data curation, J.E. and L.B.; writing—original draft preparation, J.E., L.B. and
M.B.; writing—review and editing, J.E., M.B., R.H. and M.F.Z; visualization, J.E., L.B. and M.B.;
supervision, R.H., J.E., M.B. and M.F.Z.; project administration, J.E., R.H. and M.F.Z.; funding
acquisition, J.E., R.H., M.B. and M.F.Z. All authors have read and agreed to the published version of
the manuscript.
Funding: This project has received funding from the European Union’s Horizon 2020 research and
innovation program under grant agreement number 869931 (COGNIPLANT).
Data Availability Statement: The reference data presented in this study are available from the
corresponding author on request.
Conflicts of Interest: The authors have no conflict of interest to declare.

Abbreviations
The following abbreviations are used in this manuscript:

DOF degree of freedom

EMA experimental modal analysis
FRAC frequency response assurance criterion
FRF frequency response function
GA genetic algorithm
LGS linear guiding system
LSCE least squares complex exponential
LSCF least squares complex frequency-domain
LSRF least squares rational function
MAC modal assurance criterion
MDOF multi-degree-of-freedom
ME mounting element
NDD natural damping difference
NFD natural frequency difference
NI National Instruments
OMA operational modal analysis
SSI stochastic subspace identification
SVD singular value decomposition
WPT workpiece table

References
1. Lau, J.; Lanslots, J.; Peeters, B.; van der Auweraer, H. Automatic Modal Analysis—Myth or Reality? In Proceedings of the 25th
International Modal Analysis Conference, Orlando, FL, USA, 19–22 February 2007.
2. Verboven, P. Frequency-Domain System Identification for Modal Analysis. Ph.D. Thesis, Vrije Universiteit Brussel, Brussels,
Belgium, 2019.
3. Ellinger, J.; Zaeh, M.F. Automated Identification of Linear Machine Tool Model Parameters Using Global Sensitivity Analysis.
Machines 2022, 10, 535. [CrossRef]
Appl. Sci. 2023, 13, 949 16 of 16

4. Reynders, E. System Identification Methods for (Operational) Modal Analysis: Review and Comparison. Arch. Comput. Methods
Eng. 2012, 19, 51–124. [CrossRef]
5. Ewins, D. Modal Testing: Theory, Practice and Application, 2nd ed.; Mechanical Engineering Research Studies; Research Studies
Press: Baldock, UK, 2000; Volume 10.
6. Altintas, Y. Manufacturing Automation: Metal Cutting Mechanics, Machine Tool Vibrations, and CNC Design; Cambridge University
Press: Cambridge, UK, 2012.
7. Ozdemir, A.; Gumussoy, S. Transfer Function Estimation in System Identification Toolbox via Vector Fitting. IFAC-PapersOnLine
2017, 50, 6232–6237. [CrossRef]
8. Vold, H.; Kundrat, J.; Rocklin, T.; Russell, R. A Multi-Input Modal Estimation Algorithm for Mini-Computers; SAE Technical Paper
Series; SAE International: Warrendale, PA, USA, 1982; . [CrossRef]
9. Peeters, B.; van der Auweraer, H.; Guillaume, P.; Leuridan, J. The PolyMAX Frequency-Domain Method: A New Standard for
Modal Parameter Estimation? Shock Vib. 2004, 11, 395–409. [CrossRef]
10. van Overschee, P.; de Moor, B. Subspace Identification for Linear Systems; Springer: Boston, MA, USA, 1996. [CrossRef]
11. Scionti, M.; Lanslots, J. Stabilisation Diagrams: Pole Identification Using Fuzzy Clustering Techniques. Adv. Eng. Softw. 2005,
36, 768–779. [CrossRef]
12. Reynders, E.; Houbrechts, J.; de Roeck, G. Fully Automated (Operational) Modal Analysis. Mech. Syst. Signal Process. 2012,
29, 228–250. [CrossRef]
13. Scionti, M.; Lanslots, J.; Goethals, I.; Vecchio, A.; van der Auweraer, H.; Peeters, B.; de Moor, B. Tools to Improve Detection
of Structural Changes from In-Flight Flutter Data. In Proceedings of the 8th International Conference on Recent Advances in
Structural Dynamics, Southampton, UK, 14–16 July 2003.
14. van der Auweraer, H.; Peeters, B. Discriminating Physical Poles from Mathematical Poles in High Order Systems: Use and
Automation of the Stabilization Diagram. In Proceedings of the 21st IEEE Instrumentation and Measurement Technology
Conference (IEEE Cat. No.04CH37510), Como, Italy, 18–20 May 2004; pp. 2193–2198. [CrossRef]
15. Neu, E.; Janser, F.; Khatibi, A.; Orifici, C. Fully Automated Operational Modal Analysis Using Multi-Stage Clustering. Mech. Syst.
Signal Process. 2017, 84, 308–323. [CrossRef]
16. Mugnaini, V.; Zanotti, L.; Civera, M. A Machine Learning Approach for Automatic Operational Modal Analysis. Mech. Syst.
Signal Process. 2022, 170, 108813. [CrossRef]
17. Zaletelj, K.; Bregar, T.; Gorjup, D.; Slavič, J. pyEMA . 2020. Available online: https://zenodo.org/record/4016671#.Y7vYsBVBzIU
(accessed on 26 December 2022).
18. Garnett, R. Bayesian Optimization; Cambridge University Press: Cambridge, UK, 2022.
19. Heylen, W.; Lammens, S. FRAC: A Consistent Way of Comparing Frequency Response Functions. In Identification in Engineering
Systems; Friswell, M.I., Mottershead, J., Eds.; University of Wales: Swansea, UK, 1996; pp. 48–57.
20. Allemang, R. The Modal Assurance Criterion – Twenty Years of Use and Abuse. J. Sound Vib. 2003, 37, 14–23.
21. Imamovic, N. Validation of Large Structural Dynamics Models Using Modal Test Data. Ph.D. Thesis, Imperial College of Science,
Technology & Medicine, London, UK, 1998.
22. Semm, T.; Sellemond, M.; Rebelein, C.; Zaeh, M.F. Efficient Dynamic Parameter Identification Framework for Machine Tools.
J. Manuf. Sci. Eng. 2020, 142, 081003. [CrossRef]
23. Hernandez-Vazquez, J.; Garitaonandia, I.; Fernandes, M.H.; Munoa, J.; Lacalle, L.N. A Consistent Procedure Using Response
Surface Methodology to Identify Stiffness Properties of Connections in Machine Tools. Materials 2018, 11, 1220. [CrossRef]
[PubMed]
24. Vacher, P.; Jacquier, B.; Bucharles, A. Extensions of the MAC Criterion to Complex Modes. In Proceedings of the 24th Interna-
tional Conference on Noise and Vibration Engineering, 2010, Proceedings of ISMA2010 Including USD2010, Leuven, Belgium,
20–22 September 2010; pp. 2713–2726.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.