Pose-Dependent Modal Analysis of Industrial Robots

In this study, I examine how the modal parameters of six-joint industrial robots depend on their spatial pose. I combine a pose-dependent multibody dynamic formulation with experimental modal testing to identify the physical mechanisms that cause natural frequencies, damping ratios, and mode shapes to vary across the workspace. The motivation is practical: industrial robots used for high-speed milling, grinding, polishing, and other material-removal tasks are often limited by low structural stiffness and configuration-dependent dynamics. I derive the mass, stiffness, and damping matrices as explicit functions of the generalized coordinates, and I then verify the predicted trends through hammer-impact modal experiments on a six-joint industrial robot under twelve representative poses. The results show that low-order modes are highly sensitive to pose variation, while high-order modes are comparatively stable. This paper provides a theoretical and experimental basis for stability prediction, pose selection, and model parameter identification in robotic machining systems.

1. Introduction

Industrial robots have become increasingly important in aerospace, automotive, and general manufacturing because they offer large workspaces, high flexibility, and favorable cost-effectiveness. However, the open-chain, cantilever-like structure of serial industrial robots leads to relatively low stiffness. During machining, this low stiffness can cause chatter, excessive tool-tip displacement, poor surface finish, and reduced service life. The dynamic characteristics of the robot system, including natural frequency, damping ratio, and mode shape, are therefore critical to machining stability and accuracy.

For serial industrial robots, the mass distribution and equivalent stiffness change with the robot pose. The Jacobian matrix and inertia tensor are both functions of the joint coordinates, so the system matrices are pose dependent. This means that modal parameters are not fixed properties of the robot; they vary continuously over the workspace. Understanding this pose dependence is essential for selecting stable machining configurations, optimizing tool paths, tuning process parameters, and designing active vibration control strategies.

Previous studies have addressed related problems through finite element analysis, multibody dynamics, stiffness mapping, frequency response coupling, and machining stability analysis. These works have provided important foundations for pose-dependent dynamic modeling of industrial robots. Nevertheless, a systematic experimental characterization of modal parameter variation for six-joint serial industrial robots under common machining poses remains insufficient. In addition, real-world effects such as joint flexibility, connection clearance, friction, servo locking, and assembly boundaries are difficult to model completely. Therefore, experimental modal analysis is still necessary to obtain reliable modal parameters under actual working conditions.

In this paper, I address these gaps by establishing a Lagrangian multibody dynamic model of a six-joint industrial robot, deriving the pose-dependent expressions for the mass, stiffness, and damping matrices, and performing hammer-impact modal tests with a commercial modal acquisition and analysis system. I compare the theoretical trends with the measured frequency variation and summarize the main laws of pose dependence. The work is intended to support subsequent stability analysis, working-pose selection, and model updating for industrial robots.

2. Pose-Dependent Multibody Dynamic Modeling

2.1 Kinematic Description and Denavit-Hartenberg Transformation

I use the modified Denavit-Hartenberg method to establish the coordinate frames of the robot links. For an n-degree-of-freedom serial robot, the homogeneous transformation matrix from link frame i to link frame i−1 is expressed as

$$
{}^{i-1}\boldsymbol{T}_i =
\begin{bmatrix}
\cos\theta_i & -\sin\theta_i\cos\alpha_i & \sin\theta_i\sin\alpha_i & a_i\cos\theta_i \\
\sin\theta_i & \cos\theta_i\cos\alpha_i & -\cos\theta_i\sin\alpha_i & a_i\sin\theta_i \\
0 & \sin\alpha_i & \cos\alpha_i & d_i \\
0 & 0 & 0 & 1
\end{bmatrix}
$$

where \(\theta_i\) is the joint angle, \(d_i\) is the link offset, \(a_i\) is the link length, and \(\alpha_i\) is the link twist. The end-effector velocity is related to the joint velocity by the Jacobian matrix:

$$
\begin{bmatrix}
\boldsymbol{v} \\
\boldsymbol{\omega}
\end{bmatrix}
=
\boldsymbol{J}(\boldsymbol{q})\dot{\boldsymbol{q}}
$$

The Jacobian matrix \(\boldsymbol{J}(\boldsymbol{q})\) is a nonlinear function of the generalized coordinates \(\boldsymbol{q}\). I define the generalized coordinates, velocities, and accelerations as

$$
\boldsymbol{q} = [q_1, q_2, \ldots, q_n]^T
$$

$$
\dot{\boldsymbol{q}} = [\dot{q}_1, \dot{q}_2, \ldots, \dot{q}_n]^T,
\qquad
\ddot{\boldsymbol{q}} = [\ddot{q}_1, \ddot{q}_2, \ldots, \ddot{q}_n]^T
$$

2.2 Lagrangian Formulation

The Lagrangian of the robot system is defined as the difference between kinetic energy and potential energy:

$$
L = T – V
$$

The linear velocity of the center of mass of link i is written as

$$
\boldsymbol{v}_i = \boldsymbol{J}_{i,\text{linear}}\dot{\boldsymbol{q}}
$$

The translational kinetic energy of a single link is

$$
T_{\text{linear},i}
=
\frac{1}{2}m_i\boldsymbol{v}_i^T\boldsymbol{v}_i
=
\frac{1}{2}m_i\dot{\boldsymbol{q}}^T
\boldsymbol{J}_{i,\text{linear}}^T
\boldsymbol{J}_{i,\text{linear}}
\dot{\boldsymbol{q}}
$$

The angular velocity of link i is related to the generalized velocity by

$$
\boldsymbol{\omega}_i = \boldsymbol{J}_{i,\text{angular}}\dot{\boldsymbol{q}}
$$

The rotational kinetic energy of a single link is

$$
T_{\text{rotational},i}
=
\frac{1}{2}\boldsymbol{\omega}_i^T\boldsymbol{I}_i\boldsymbol{\omega}_i
=
\frac{1}{2}\dot{\boldsymbol{q}}^T
\boldsymbol{J}_{i,\text{angular}}^T
\boldsymbol{I}_i
\boldsymbol{J}_{i,\text{angular}}
\dot{\boldsymbol{q}}
$$

where \(\boldsymbol{I}_i\) is the inertia tensor of link i expressed in the world coordinate frame. The total kinetic energy of the system can be written in quadratic form:

$$
T = \frac{1}{2}\dot{\boldsymbol{q}}^T\boldsymbol{M}(\boldsymbol{q})\dot{\boldsymbol{q}}
$$

The pose-dependent mass matrix is therefore

$$
\boldsymbol{M}(\boldsymbol{q})
=
\sum_{i=1}^{n}
\left[
m_i\boldsymbol{J}_{i,\text{linear}}^T\boldsymbol{J}_{i,\text{linear}}
+
\boldsymbol{J}_{i,\text{angular}}^T\boldsymbol{I}_i\boldsymbol{J}_{i,\text{angular}}
\right]
$$

Because both the Jacobian matrix \(\boldsymbol{J}(\boldsymbol{q})\) and the inertia tensor \(\boldsymbol{I}_i(\boldsymbol{q})\) depend on the pose \(\boldsymbol{q}\), the mass matrix \(\boldsymbol{M}(\boldsymbol{q})\) has explicit pose dependence. This is the first major source of configuration-dependent dynamics in industrial robots.

The gravitational potential energy is the sum of the potential energies of all link centers of mass:

$$
V_g = \sum_{i=1}^{n}m_i\boldsymbol{g}^T\boldsymbol{p}_{c,i}(\boldsymbol{q})
$$

where \(\boldsymbol{p}_{c,i}(\boldsymbol{q})\) is the position vector of the center of mass of link i in the world frame.

2.3 Joint Flexibility and Cartesian Stiffness Mapping

I consider elastic deformation mainly from the joint reducers. The joint stiffness matrix is defined as

$$
\boldsymbol{K}_{\theta}
=
\operatorname{diag}(k_{\theta1}, k_{\theta2}, \ldots, k_{\theta n})
$$

The elastic potential energy in joint space is

$$
V_{\text{elastic}}
=
\frac{1}{2}\delta\boldsymbol{q}^T
\boldsymbol{K}_{\theta}
\delta\boldsymbol{q}
$$

Using the Jacobian matrix, the joint-space stiffness is mapped to Cartesian space as

$$
\boldsymbol{K}_{\text{Cartesian}}(\boldsymbol{q})
=
\boldsymbol{J}^{-T}(\boldsymbol{q})
\boldsymbol{K}_{\theta}
\boldsymbol{J}^{-1}(\boldsymbol{q})
$$

This local small-perturbation stiffness mapping is valid near a non-singular equilibrium pose. It requires that the Jacobian matrix remains full rank and invertible in the selected task space, that the joint-space and task-space dimensions match, and that the pose remains away from singular configurations. Under these conditions, the Cartesian stiffness matrix depends explicitly on the robot pose. This dependence is a major reason for the anisotropic and pose-dependent end-effector stiffness of industrial robots.

2.4 Damping and the Pose-Dependent Dynamic Equation

I adopt the Rayleigh damping model and assume that the damping matrix is a linear combination of the mass and stiffness matrices:

$$
\boldsymbol{B}(\boldsymbol{q})
=
\alpha\boldsymbol{M}(\boldsymbol{q})
+
\beta\boldsymbol{K}(\boldsymbol{q})
$$

Because \(\boldsymbol{M}(\boldsymbol{q})\) and \(\boldsymbol{K}(\boldsymbol{q})\) both depend on pose, the damping matrix \(\boldsymbol{B}(\boldsymbol{q})\) is also pose dependent. The dynamic equation of the robot with joint flexibility is

$$
\boldsymbol{M}(\boldsymbol{q})\ddot{\boldsymbol{q}}
+
\boldsymbol{B}(\boldsymbol{q})\dot{\boldsymbol{q}}
+
\boldsymbol{K}(\boldsymbol{q})\boldsymbol{q}
=
\boldsymbol{\tau}
$$

This equation reveals the essential feature of the system: the matrices \(\boldsymbol{M}\), \(\boldsymbol{B}\), and \(\boldsymbol{K}\) are nonlinear functions of the generalized coordinates \(\boldsymbol{q}\). This mathematically proves the pose dependence of the dynamic characteristics of industrial robots.

2.5 Eigenvalue Problem and Sensitivity

Neglecting damping, the free-vibration equation is

$$
\boldsymbol{M}(\boldsymbol{q})\ddot{\boldsymbol{q}}
+
\boldsymbol{K}(\boldsymbol{q})\boldsymbol{q}
=
0
$$

Assuming a harmonic solution \(\boldsymbol{q} = \boldsymbol{\Phi}e^{j\omega t}\), I obtain the generalized eigenvalue problem:

$$
\boldsymbol{K}(\boldsymbol{q})\boldsymbol{\Phi}_r(\boldsymbol{q})
=
\omega_r^2(\boldsymbol{q})
\boldsymbol{M}(\boldsymbol{q})\boldsymbol{\Phi}_r(\boldsymbol{q})
$$

For any mode r, the pose-dependent natural frequency is

$$
\omega_r(\boldsymbol{q})
=
\sqrt{
\frac{
\boldsymbol{\Phi}_r^T(\boldsymbol{q})
\boldsymbol{K}(\boldsymbol{q})
\boldsymbol{\Phi}_r(\boldsymbol{q})
}{
\boldsymbol{\Phi}_r^T(\boldsymbol{q})
\boldsymbol{M}(\boldsymbol{q})
\boldsymbol{\Phi}_r(\boldsymbol{q})
}
}
$$

This expression shows that the natural frequency is determined by the ratio of the modal stiffness to the modal mass. When the robot pose changes, two effects occur simultaneously. First, the force transmission path changes, and the equivalent stiffness changes. Second, the relative positions of the links change, and the inertia matrix components change. The coupling between these two effects determines the final frequency variation.

The first-order sensitivity of the modal frequency with respect to a joint angle is

$$
S_{r,i}(\boldsymbol{q})
=
\frac{\partial\omega_r(\boldsymbol{q})}{\partial q_i}
=
\frac{1}{2\omega_r(\boldsymbol{q})}
\frac{\partial}{\partial q_i}
\left(
\frac{
\boldsymbol{\Phi}_r^T\boldsymbol{K}\boldsymbol{\Phi}_r
}{
\boldsymbol{\Phi}_r^T\boldsymbol{M}\boldsymbol{\Phi}_r
}
\right)
$$

The second-order sensitivity is

$$
C_{r,ij}(\boldsymbol{q})
=
\frac{\partial^2\omega_r(\boldsymbol{q})}{\partial q_i\partial q_j}
$$

Sensitivity analysis can identify the joints that most strongly influence modal frequencies, predict frequency trends, and provide a basis for pose optimization. The modal damping ratio is also pose dependent:

$$
\xi_r(\boldsymbol{q})
=
\frac{
\boldsymbol{\Phi}_r^T(\boldsymbol{q})
\boldsymbol{B}(\boldsymbol{q})
\boldsymbol{\Phi}_r(\boldsymbol{q})
}{
2\sqrt{
\boldsymbol{\Phi}_r^T(\boldsymbol{q})
\boldsymbol{K}(\boldsymbol{q})
\boldsymbol{\Phi}_r(\boldsymbol{q})
\cdot
\boldsymbol{\Phi}_r^T(\boldsymbol{q})
\boldsymbol{M}(\boldsymbol{q})
\boldsymbol{\Phi}_r(\boldsymbol{q})
}
}
$$

2.6 Physical Mechanism of Pose Dependence

For the center of mass of link i and the j-th rotational joint, the linear velocity Jacobian column can be written as

$$
\boldsymbol{J}_{v,i}^{(j)}(\boldsymbol{q})
=
\boldsymbol{z}_{j-1}(\boldsymbol{q})
\times
\left[
\boldsymbol{p}_{c,i}(\boldsymbol{q})
–
\boldsymbol{p}_{j-1}(\boldsymbol{q})
\right],
\qquad j \le i
$$

The term \(\boldsymbol{p}_{c,i} – \boldsymbol{p}_{j-1}\) represents the geometric moment arm from the joint axis to the link center of mass. In an extended pose, the distance between the distal link center of mass and the main joint axes increases. The corresponding components of \(\boldsymbol{J}_{v,i}\) become larger, and the projection of the link inertia tensor in the base frame also changes. These components increase the contribution of the inertia terms to the low-order global swing modes, so the equivalent modal mass \(m_{r,\text{eq}} = \boldsymbol{\Phi}_r^T\boldsymbol{M}\boldsymbol{\Phi}_r\) tends to increase. In a compact pose, the long moment arms are reduced, and the equivalent modal mass of the low-order modes decreases.

For stiffness, the joint flexibility mapped to task space can be written in compliance form. For a unit direction \(\boldsymbol{u}\) in task space, the directional compliance is

$$
c_u(\boldsymbol{q})
=
\left[
\boldsymbol{J}^T(\boldsymbol{q})\boldsymbol{u}
\right]^T
\boldsymbol{K}_{\theta}^{-1}
\left[
\boldsymbol{J}^T(\boldsymbol{q})\boldsymbol{u}
\right]
$$

In an extended pose, the long moment arms increase the amplitude of \(\boldsymbol{J}^T\boldsymbol{u}\) in the main bending directions. The displacement amplification of joint flexibility is enhanced, so the directional compliance increases and the corresponding directional stiffness decreases. In a compact pose, the force transmission path is shortened, the displacement amplification effect of joint flexibility is weakened, and the equivalent stiffness increases.

According to the frequency expression, the r-th natural frequency satisfies

$$
\omega_r^2(\boldsymbol{q})
=
\frac{
\boldsymbol{\Phi}_r^T(\boldsymbol{q})
\boldsymbol{K}(\boldsymbol{q})
\boldsymbol{\Phi}_r(\boldsymbol{q})
}{
\boldsymbol{\Phi}_r^T(\boldsymbol{q})
\boldsymbol{M}(\boldsymbol{q})
\boldsymbol{\Phi}_r(\boldsymbol{q})
}
=
\frac{k_{r,\text{eq}}(\boldsymbol{q})}{m_{r,\text{eq}}(\boldsymbol{q})}
$$

where \(k_{r,\text{eq}} = \boldsymbol{\Phi}_r^T\boldsymbol{K}\boldsymbol{\Phi}_r\) is the equivalent modal stiffness. Therefore, I can make the following prior judgment. An extended pose increases the equivalent modal mass and decreases the equivalent modal stiffness in low-order mode directions, so the ratio \(k_{r,\text{eq}} / m_{r,\text{eq}}\) decreases and the low-order natural frequencies decrease. A compact pose decreases the equivalent modal mass and increases the equivalent modal stiffness, so the low-order natural frequencies increase. Both effects act in the same direction for low-order modes, which makes them highly sensitive to overall pose variation. For high-order modes, local bending and torsion components are more significant, and the projection of the global moment-arm variation onto their mode directions is weaker. Therefore, high-order modes are less sensitive to extended and compact pose changes.

Summary of pose-dependent system matrices and physical causes
System matrix Pose dependence Main physical cause Expected effect on low-order modes
Mass matrix \(\boldsymbol{M}(\boldsymbol{q})\) Nonlinear function of joint coordinates Jacobian matrix and inertia tensor change with link geometry Extended pose increases equivalent modal mass
Stiffness matrix \(\boldsymbol{K}(\boldsymbol{q})\) Nonlinear function of joint coordinates Joint flexibility mapped through Jacobian inverse Extended pose decreases equivalent modal stiffness
Damping matrix \(\boldsymbol{B}(\boldsymbol{q})\) Rayleigh combination of \(\boldsymbol{M}\) and \(\boldsymbol{K}\) Pose dependence inherited from mass and stiffness Damping ratio changes with configuration
Natural frequency \(\omega_r(\boldsymbol{q})\) Ratio of modal stiffness to modal mass Coupled mass and stiffness variation Compact pose generally increases low-order frequency
Damping ratio \(\xi_r(\boldsymbol{q})\) Modal projection of \(\boldsymbol{B}\) Energy dissipation path changes with pose Pose-dependent but often less monotonic

3. Experimental Modal Analysis

3.1 Test Object and Measurement System

I selected a six-joint serial industrial robot for experimental modal testing. The robot is representative of a class of industrial robots used for milling, grinding, and polishing tasks. Its joint motion parameters are listed in the table below. The maximum joint travels and maximum joint speeds are given for each axis.

Joint motion parameters of the tested six-joint industrial robot
Joint Maximum travel (deg) Maximum speed (deg/s)
1 ±185 100
2 −65 to +88 90
3 −165 to +80 90
4 ±360 140
5 ±125 110
6 ±360 180

The modal test system consists of an impact hammer, accelerometers, a data acquisition front end, and modal analysis software. I used a commercial modal testing platform for data processing and parameter identification. The software supports impact testing, frequency response function estimation, coherence checking, stabilization diagram construction, and polyreference modal identification.

I built a simplified geometric model of the industrial robot in the modal software, including the topology and geometric dimensions of the links. Based on the structural features of the robot, I placed 52 measurement points at key locations. Each measurement point was used to acquire acceleration responses in three orthogonal directions. The measurement point layout followed three principles: covering all major links, identifying the dominant mode shapes in the frequency range of interest, and avoiding nodal positions where the response is close to zero.

3.2 Hammer Impact Test Procedure

I adopted a fixed-response, moving-excitation strategy. A three-axis accelerometer was mounted on the end flange as the reference sensor. I then impacted each measurement point sequentially in the X, Y, and Z directions. This strategy reduces the number of sensors, lowers mass-loading effects, and ensures consistency of the frequency response functions for complete mode shape identification.

Experimental parameter settings for hammer impact modal testing
Parameter Setting Purpose
Sampling frequency 2048 Hz Satisfies the sampling theorem
Frequency resolution 0.5 Hz Analysis bandwidth divided by number of spectral lines
Analysis bandwidth 0–512 Hz Covers the frequency range of interest
Number of averages 3 Improves signal-to-noise ratio
Force window Rectangular Preserves impact signal integrity
Exponential window Decay constant = 100% Reduces truncation leakage
Trigger mode Force signal trigger Trigger level = 1%
Overload rejection Enabled Rejects double impacts and abnormal data

The test procedure was as follows. First, I calibrated the sensor sensitivity and checked channel consistency. Second, I adjusted the robot to the target pose and kept the servo locked. Third, I impacted the measurement points sequentially and acquired the frequency response functions. At each pose, I repeated the acquisition according to the averaging settings. I then screened the data using the frequency response function shape and the coherence function. When a resonance peak was unclear or the coherence function was below 0.8, I discarded the measurement and repeated the test. Finally, I used the valid averaged results that satisfied the coherence requirement for modal parameter identification. This procedure reduced the influence of single-impact variability and local contact differences.

3.3 Modal Parameter Identification and Data Quality

I used the PolyMAX algorithm for modal parameter identification. This algorithm is based on the polyreference least-squares complex frequency-domain method. It provides clear stabilization diagrams and is effective at distinguishing physical poles from computational poles. Its high accuracy makes it suitable for dense modes and highly damped systems.

I first reviewed the frequency response function and coherence function. The main checks were whether the curve shape was smooth, whether the resonance peaks were clear, and whether the coherence function was sufficiently high. The coherence function measures the linear correlation between excitation and response, with values between 0 and 1. Near a resonance peak, the coherence value should be close to 1 in an ideal measurement.

Coherence-based data quality criteria
Coherence value Data quality Action
Greater than 0.9 Excellent Accept measurement
0.8 to 0.9 Good Accept with review
Less than 0.8 Insufficient Repeat measurement

In the frequency range of interest, the coherence function was generally greater than 0.8. The resonance peaks were clearly identifiable, and the separation between adjacent modes was sufficient. The low-frequency band below 100 Hz showed high measurement quality. After obtaining the frequency response function data set, I entered the modal data selection stage and then performed PolyMAX modal analysis. I selected the bandwidth of interest and used the stabilization diagram to calculate modal parameters. From the stable poles, I extracted the natural frequency \(f_n\), damping ratio \(\xi\), and mode shape \(\boldsymbol{\Phi}\).

4. Pose Selection and Experimental Results

4.1 Pose Sampling Set

The pose points were selected from the workspace commonly covered by milling, grinding, and similar machining tasks. I selected twelve poses for modal testing. The sampling points cover far-extended, medium forward, compact folded, elbow-folded, wrist-pitching, and circumferential perturbation configurations. They include common machining poses and the nearest and farthest end-effector machining points. The pose set forms a comparison among extended, intermediate, and compact configurations.

Pose sampling points used in the modal verification experiments
Pose Joint angles (deg) Pose description
P1 (0, −45, 0, 0, 0, 0) Far-extended baseline
P2 (0, −10, −45, 0, −90, 0) Compact folded-elbow baseline
P3 (0, −35, −15, 45, −40, −35) Medium forward transition
P4 (0, −25, −32, −60, −40, 0) Medium extension with elbow deflection
P5 (0, −10, −45, 0, −145, 0) Medium deflection with deep wrist pitch
P6 (10, −50, 20, 60, −200, 0) Far extension with large wrist pitch
P7 (−12, −35, −7, 33, −43, 28) Medium forward with slight perturbation
P8 (−7, −10, −42, 116, −90, −90) Compact with wrist upturn and circumferential difference
P9 (9, −22, −29, −0.7, −162, 10) Medium forward with extreme wrist pitch
P10 (15, 60, 86, 0, 4, 12) More compact with positive elbow angle
P11 (−5, −39, 3, 91, −22, −35) Far extension with circumferential perturbation
P12 (−10, −60, 48, 174, −40, 20) Far extension with high wrist posture

The end-effector positions corresponding to these poses provide coverage of the studied working conditions. The pose set is intended to represent the range of configurations that industrial robots commonly encounter during machining. I use this set to compare extended poses, intermediate poses, and compact poses under otherwise consistent modal testing conditions.

4.2 Experimental Modal Frequencies

The modal frequencies extracted under each pose are listed below. The first five modes are reported because they dominate the low-frequency dynamic behavior relevant to machining stability.

Experimental natural frequencies under each pose (Hz)
Pose Mode 1 Mode 2 Mode 3 Mode 4 Mode 5
P1 3.555 6.555 12.774 14.480 18.883
P2 4.580 7.249 12.085 15.801 17.353
P3 3.714 6.754 12.649 14.272 18.643
P4 4.774 6.999 13.254 14.563 18.426
P5 5.354 7.204 12.945 14.951 18.100
P6 4.387 6.168 13.468 14.734 20.221
P7 4.406 6.504 13.182 15.584 18.923
P8 5.477 7.190 13.051 14.570 19.500
P9 4.939 6.735 13.069 14.663 18.480
P10 8.232 10.584 12.269 18.053 24.704
P11 4.100 6.368 12.447 14.627 19.263
P12 3.834 5.567 13.254 14.440 20.811

The experimental results show that the natural frequencies of the industrial robot exhibit strong pose dependence. The differences between compact and extended poses are substantial. For the first mode, the frequency increases from 3.555 Hz to 8.232 Hz, corresponding to a variation rate of 131.6%. For the second mode, the frequency increases from 5.567 Hz to 10.584 Hz, corresponding to a variation rate of 90.1%. For the fourth mode, the frequency increases from 14.272 Hz to 18.053 Hz, corresponding to a variation rate of 26.5%. For the fifth mode, the frequency increases from 17.353 Hz to 24.704 Hz, corresponding to a variation rate of 42.4%. The third mode remains relatively stable, distributed between 12.085 Hz and 13.468 Hz.

Frequency variation range and rate for each mode
Mode Minimum frequency (Hz) Maximum frequency (Hz) Variation rate (%) Sensitivity level
Mode 1 3.555 8.232 131.6 Very high
Mode 2 5.567 10.584 90.1 High
Mode 3 12.085 13.468 11.4 Low
Mode 4 14.272 18.053 26.5 Moderate
Mode 5 17.353 24.704 42.4 Moderate to high

4.3 Frequency Trends and Sensitivity

The measured frequencies generally follow the rule that extended poses have lower frequencies and compact poses have higher frequencies. This trend holds for all modes except the third mode, which shows relatively small variation. The low-order modes, especially the first and second modes, are much more sensitive to pose variation than the higher-order modes. This is consistent with the theoretical prediction that pose variation changes the equivalent mass and stiffness most strongly in the directions of global low-order deformation modes.

Comparison of extended and compact pose effects on modal parameters
Quantity Extended pose Compact pose Frequency consequence
Link center-of-mass moment arm Large Small Changes inertial coupling
Equivalent modal mass Higher Lower Extended pose lowers frequency
Force transmission path Longer Shorter Changes effective stiffness
Equivalent modal stiffness Lower Higher Compact pose raises frequency
Low-order frequency Lower Higher Strong sensitivity
High-order frequency Less affected Less affected Weak to moderate sensitivity

The third mode is an exception to the general trend. Its frequency varies only slightly across the tested poses. This relative stability may be related to the local structural characteristics of the links, the joint motor assemblies, nonlinear deformation components, and the distribution of the sampled poses. I will investigate this behavior further in future work by combining detailed structural modeling with additional pose sampling.

4.4 Mechanism Interpretation of the Experimental Trends

The experimental results agree with the pose-dependent mechanism derived in the theoretical section. The first two modes are global low-order modes. Their mode shapes involve large-scale motion of the arm and wrist. The equivalent modal mass and stiffness of these modes are strongly influenced by the geometric moment arms between the joint axes and the link centers of mass. In an extended pose, these moment arms are large. The mass matrix contribution from the distal links increases, and the force transmission path becomes longer. As a result, the equivalent modal mass increases and the equivalent modal stiffness decreases. Both effects reduce the natural frequency. In a compact pose, the moment arms are shorter, the equivalent modal mass decreases, and the equivalent modal stiffness increases. Both effects raise the natural frequency.

For high-order modes, the mode shapes contain more local bending and torsion components. The projection of the global arm-length variation onto these local mode directions is weaker. Therefore, the frequency variation of high-order modes is smaller. This explains why the fourth and fifth modes vary less than the first and second modes.

The pose dependence can be summarized by the following ratio:

$$
\frac{\omega_r(\boldsymbol{q}_{\text{compact}})}{\omega_r(\boldsymbol{q}_{\text{extended}})}
=
\sqrt{
\frac{
k_{r,\text{eq}}(\boldsymbol{q}_{\text{compact}})
/
m_{r,\text{eq}}(\boldsymbol{q}_{\text{compact}})
}{
k_{r,\text{eq}}(\boldsymbol{q}_{\text{extended}})
/
m_{r,\text{eq}}(\boldsymbol{q}_{\text{extended}})
}
}
$$

For the first mode, the measured frequency ratio between the most compact and most extended poses is

$$
\frac{\omega_1(\boldsymbol{q}_{P10})}{\omega_1(\boldsymbol{q}_{P1})}
=
\frac{8.232}{3.555}
\approx 2.316
$$

For the second mode, the ratio is

$$
\frac{\omega_2(\boldsymbol{q}_{P10})}{\omega_2(\boldsymbol{q}_{P12})}
=
\frac{10.584}{5.567}
\approx 1.901
$$

These ratios indicate that the equivalent stiffness-to-mass ratio can change by a factor of more than two for the first mode. This is a critical consideration for stability analysis and process planning in industrial robots. A machining path that moves from an extended configuration to a compact configuration may experience a large shift in natural frequency, which can change the stability lobes and the risk of chatter.

Measured frequency ratios between representative compact and extended poses
Mode Compact pose value (Hz) Extended pose value (Hz) Ratio Equivalent physical interpretation
Mode 1 8.232 (P10) 3.555 (P1) 2.316 Strong reduction in modal mass and increase in modal stiffness
Mode 2 10.584 (P10) 5.567 (P12) 1.901 Moderate to strong mass-stiffness change
Mode 3 13.468 (P6) 12.085 (P2) 1.114 Weak sensitivity to global pose
Mode 4 18.053 (P10) 14.272 (P3) 1.265 Moderate local-global coupling
Mode 5 24.704 (P10) 17.353 (P2) 1.424 Moderate sensitivity with local components

4.5 Implications for Machining Stability and Pose Selection

The pose dependence of modal parameters has direct implications for machining stability. The stability lobes of a robotic machining system depend on the natural frequency and damping ratio at the tool tip. If the natural frequency changes by more than 100% across the workspace, a fixed spindle speed may be stable in one pose and unstable in another. Therefore, stability prediction for industrial robots should not rely on a single modal model. Instead, it should account for the pose-dependent frequency and damping.

For pose selection, compact poses generally provide higher low-order natural frequencies and therefore higher dynamic stiffness in the low-frequency range. However, compact poses may also reduce the reachable workspace and change the tool orientation. A practical strategy is to evaluate the modal parameters along the planned machining path and to avoid configurations where the dominant natural frequency approaches the tooth-passing frequency or its harmonics.

Practical considerations for pose selection in robotic machining
Consideration Extended pose Compact pose Recommended action
Workspace reach Large Small Balance reach against dynamic stiffness
Low-order natural frequency Low High Prefer compact poses for low-frequency chatter avoidance
Modal mass High Low Consider inertial loading during fast motion
Equivalent stiffness Low High Prefer compact poses for static deflection control
Stability lobe shift Large Large Update stability model with pose

5. Discussion

The theoretical and experimental results show that the modal behavior of six-joint industrial robots is strongly pose dependent. The mass matrix, stiffness matrix, and damping matrix are all nonlinear functions of the joint coordinates. The Jacobian matrix and inertia tensor change with pose, which changes the equivalent modal mass and stiffness. The experimental data confirm that the first two modes vary by 131.6% and 90.1%, respectively, over the tested poses. The fourth and fifth modes vary by 26.5% and 42.4%, respectively. The third mode is relatively stable.

This study has several implications for the modeling and control of industrial robots. First, a single modal model is insufficient for accurate stability prediction across the workspace. Pose-dependent modal models are needed. Second, experimental modal analysis remains essential because real-world effects such as joint clearance, friction, servo locking, and assembly boundaries are difficult to model completely. Third, sensitivity analysis can identify the joints that most strongly affect the modal frequencies. This information can be used for pose optimization and structural design.

The results also suggest that the pose dependence is not uniform across modes. Low-order modes are dominated by global arm motion, so they are highly sensitive to the geometric moment arms. High-order modes contain more local deformation, so they are less sensitive to global pose changes. This distinction is important for model reduction and for selecting the modes that must be included in a stability analysis.

In future work, I plan to combine joint stiffness and damping identification with grid-based end-effector sampling and model updating. The goal is to establish a quantitative relationship between model-predicted frequencies and measured frequencies and to extend the analysis to the full workspace and to different configurations of industrial robots. I also plan to investigate the relative stability of the third mode by considering link structure, joint motors, nonlinear deformation components, and additional pose sampling.

6. Conclusions

I have investigated the pose dependence of modal parameters in six-joint industrial robots through theoretical modeling and experimental verification. The main conclusions are as follows.

First, I established a pose-dependent dynamic model based on the Lagrangian method. I derived the analytical expressions for the pose-dependent mass matrix \(\boldsymbol{M}(\boldsymbol{q})\), stiffness matrix \(\boldsymbol{K}(\boldsymbol{q})\), and damping matrix \(\boldsymbol{B}(\boldsymbol{q})\). The system matrices are nonlinear functions of the generalized coordinates. I also clarified the non-singular, small-perturbation, and dimension-matching conditions required for mapping joint stiffness to Cartesian space.

Second, I verified the pose dependence through hammer-impact modal experiments. I used a commercial modal testing system and obtained modal parameters under multiple representative poses. The experimental results show that the natural frequencies of industrial robots vary significantly with pose. The first and second modes vary by 131.6% and 90.1%, respectively. The fourth and fifth modes vary by 26.5% and 42.4%, respectively. Low-order modes are more sensitive to pose variation than high-order modes.

Third, I explained the physical mechanism of pose dependence. In compact poses, the link centers of mass are closer to the base, the rotational inertia decreases, and the force transmission path is shorter. These effects increase the equivalent stiffness and reduce the equivalent modal mass, leading to higher natural frequencies. In extended poses, the cantilever structure has a larger rotational inertia and lower stiffness, leading to lower natural frequencies.

The pose-dependent modal behavior identified in this study provides a theoretical and experimental basis for stability analysis, working-pose selection, and model parameter identification of industrial robots. The results can help improve machining accuracy and stability in robotic milling, grinding, and polishing applications. Future work will focus on quantitative model updating, full-workspace modal prediction, and extension to different configurations of industrial robots.

Appendix: Summary of Key Quantities and Experimental Data

Key symbols used in the pose-dependent dynamic formulation
Symbol Meaning Pose dependence
\(\boldsymbol{q}\) Generalized coordinates Independent variable
\(\boldsymbol{J}(\boldsymbol{q})\) Jacobian matrix Nonlinear function of pose
\(\boldsymbol{M}(\boldsymbol{q})\) Mass matrix Depends on Jacobian and inertia tensor
\(\boldsymbol{K}(\boldsymbol{q})\) Stiffness matrix Depends on Jacobian inverse and joint stiffness
\(\boldsymbol{B}(\boldsymbol{q})\) Damping matrix Rayleigh combination of mass and stiffness
\(\boldsymbol{\Phi}_r(\boldsymbol{q})\) Mode shape of mode r Depends on pose
\(\omega_r(\boldsymbol{q})\) Natural frequency of mode r Ratio of modal stiffness to modal mass
\(\xi_r(\boldsymbol{q})\) Damping ratio of mode r Modal projection of damping matrix
\(k_{r,\text{eq}}\) Equivalent modal stiffness Pose dependent
\(m_{r,\text{eq}}\) Equivalent modal mass Pose dependent
Measured first-mode and second-mode frequencies for all poses
Pose Mode 1 (Hz) Mode 2 (Hz) Pose class
P1 3.555 6.555 Far extended
P2 4.580 7.249 Compact
P3 3.714 6.754 Medium forward
P4 4.774 6.999 Medium extension
P5 5.354 7.204 Medium deflection
P6 4.387 6.168 Far extended
P7 4.406 6.504 Medium forward
P8 5.477 7.190 Compact
P9 4.939 6.735 Medium forward
P10 8.232 10.584 Compact
P11 4.100 6.368 Far extended
P12 3.834 5.567 Far extended
Qualitative ranking of pose sensitivity by mode
Mode Sensitivity to pose Dominant deformation character Expected machining impact
Mode 1 Very high Global arm swing Large stability lobe shift
Mode 2 High Global arm and wrist motion Significant stability change
Mode 3 Low Local or mixed deformation Small frequency shift
Mode 4 Moderate Mixed local and global Moderate stability change
Mode 5 Moderate to high Local bending and torsion Moderate stability change

Overall, the results demonstrate that pose-dependent modal analysis is essential for understanding the dynamic behavior of industrial robots. The combination of Lagrangian modeling and experimental modal testing provides a reliable framework for predicting and interpreting frequency variation. The findings can support the design of pose-dependent stability models, the selection of favorable machining configurations, and the development of vibration control strategies for industrial robots.

Scroll to Top