simpack a friedrichshafener strasse 1 82205 ilching …€¦ · · 2014-03-20simpack a...

SIMPACK NewsSIMPACK NewsSIMPACK AG, Friedrichshafener Strasse 1, 82205 Gilching, Germany

SIMPACK scripts are now available for easily defining, running and analyzing the thousands of load cases required for proper dimen-sioning of components and fulfilling certification requirements for wind turbines. The load calculation process can be used with any SIMPACK wind turbine model, i.e., from simple models used for initial concept studies and certification purposes.... See page 19

The tire-road interface is one of the most important elements to consider when analyzing vehicle dynamics. This SIMPACK News issue features three of the most important tire models used today within multi-body simulation. See pages 2–15

Comparison of Simulation and Measurements from On-Track Tests for Model ValidationThe rail vehicle approval process today is costly and time-consuming. "Virtual testing" using multi-body simulations is one approach to reduce these costs. During the European research project DynoTRAIN, one of the... See page 22

Load Calculations for Wind Turbines

SIMPACK's 3rd Party Tire Products

MARCH 2014

26 CUSTOMER APPLICATIONKinematics Analysis and Design Optimization of Semi-Active Suspension for a Light Bus

19 SOFTWARELoad Calculations for Wind Turbines

06 3RD PARTY PRODUCTTNO’s MF-Tyre / MF-Swift and the Delft-Tyre Toolchain

22 CUSTOMER APPLICATIONComparison of Simulation and Measurements from On-Track Tests for Model Validation

10 3RD PARTY PRODUCTFTire: High-End Tire Model for Vehicle Simulation in SIMPACK

16 CUSTOMER APPLICATIONMusculoskeletal Model of Bicycle Pedaling

33 CUSTOMER APPLICATIONModeling, Simulation and Dynamic Analyses of a Closed Single-Track Vehicle

02 3RD PARTY PRODUCTTire Modeling from Structural Analysis to Real-Time Applications

Ready for Print - 20.02.2014

1. Tread layer

2. Cap ply

3. Steell cord layer #1

4. Steell cord layer #2

5. Rubber/sidewall/bead support layer

6. Carcass layer

7. Innerliner

4

7

2

1

6

5

5

5

5

© 2009 Goodyear Dunlop

3

2 | SIMPACK News | March 2014 SIMPACK News | March 2014 | 3

Axel Gallrein, Manfred Bäcker, Michael Burger, Fraunhofer-Institute ITWM | 3RD PARTY PRODUCT3RD PARTY PRODUCT | Axel Gallrein, Manfred Bäcker, Michael Burger, Fraunhofer-Institute ITWM

sidewall

Tire Modeling from Structural Analysis to Real-Time Applications

Fig. 1: Mass distribution in tire and cross section

INTRODUCTIONAll commercially available tire models — in-cluding CDTire — were developed at a time when available computer hardware deliv-ered only a fraction of the computational performance available today. Simplifications in the modeling process were accepted in order to ensure the applicability of the mod-els in productive development processes. Due to these simplifications, the application range and accuracy of the load prediction was limited. Motivated by unexploited po-tential in the virtual development process, Fraunhofer ITWM developed a full 3D structural tire model to extend both the ap-plication range and the achievable accuracy of full vehicle scenarios. The new CDTire/3D additionally opens the door to extended tire parameterization strat-egies: local geometry-based tire measure-ments can be used to reduce the number of tests needed to a minimum. CDTire/3D also enables the parameterization of tire types and sizes where parameterization failed in the past due to the non-availability of suit-able measurements. CDTire/3D is available with SIMPACK 9.4.

MODELING CONCEPT OF CDTIRE/3DThe basic concept of this modeling approach is that the local deformation behavior of

Fig. 2: Triangle pairs used for cell edge bending (left) and around diagonals of a cell (right) Fig. 4: Functional component layers of a tire

Fig. 3: Finite difference stencil of isotropic bending with unit deformation (left) and example of reaction forces to discrete deformation of belt center point (right)

a real tire should be identical to that of an MBD tire model. Consequently, the model must have a detailed, materialized shell representation of sidewall and belt to fea-ture the deformation behavior of the load bearing structure. With this, one can feature both in-plane and out-of-plane (transversal) deformation behavior.

The density properties of the shell are rep-resented by discrete mass points with every mass point having three dynamic degrees of freedom (see Fig. 1).The elastic properties of the shell are real-ized by an anisotropic elastic membrane part and adaptation of the Kirchhoff-Love hypothesis for bending. The bending is implemented around all circumferential and lateral edges, and around two diagonals of each cell. To constitute the bending laws, the 4-point cells will be divided into their elementary triangles. Bending of adjacent cells relative to the edge is split into bending of two pairs of triangles. The same is done for bending around the diagonals of a cell, see Fig. 2 and Fig. 3.

COMPONENT PROPERTY LAYERSThe anisotropy of the tire is a direct conse-quence of the tire structure. The physical tire is built from different component layers, e.g., inner liner, carcass, steel belt layers,

cap plies, tread etc., with most of these being reinforced by synthetic cords or steel wires. Each reinforcement layer introduces directionally dependent material properties, the orientation of the cords relative to the circumferential direction is specified with an angle. All the characteristic component lay-ers described above have a separate repre-sentation in the model. The main advantage of this de-scription is that the model is completely configurable. One can, for example, mod-el an arbitrary number of steel belt layers. Every steel belt layer can have a specific belt angle and specific local stiffness properties.

This is important for modeling truck tires, which can have a varying number of steel belt layers and varying angles compared to passenger car tires, see Fig. 4.

GEOMETRY BASED MODELINGThe first key entity is the cross section. The modeling starts by specifying the discrete

mass distribution of the non-inflated cross section. Then, the mass-points (nodes) and the related cells must be

defined. Based on the nodes and cells in the cross section, one can define every compo-nent layer’s local cross section parameters,

CAE-based methods are continuously accelerating the vehicle development pro-cess. Virtual vehicle models drive over the digitized road surfaces of OEM’s proving grounds, steered by virtual driver models. One of the key components of the load transfer process is the tire. The quality of the underlying tire model is crucial for the accuracy and reliability of virtual load prediction scenarios.

Fraunhofer ITWM’s new tire models CDTire/3D and CDTire/Realtime have been implemented in SIMPACK to improve the ac-curacy and ease of parameterization for large deformations and lateral dynamics, as well as supporting comfort and durability applications in realtime systems.

belt

rim

which are related to nodes or to cells. The way to do this follows a global-to-local concept.

In the global-to-local concept, the user first specifies a global

parameter which can be seen as the average

of the cross section, and then specifies the local param-eters by defining weights relative to

the global param-eter. The specification

of weights is optional.For the reinforced lay-

ers (carcass, steel cord and cap plies), one can specify the local and

component-specific stiffness and damping parameters, as well as local pre-stresses. This is done by specifying local reduction factors for the stress-free characteristic lengths of the reinforcement material (syn-thetic or steel wires).

CONTACT FORMULATIONThe contact model is a brush-type contact model. The single bristles act as sensors; they detect road contact and their deforma-tion results in a force transferred to the tire structure. The bristle tips can stick or slide. The number of sensors and their lengths can be specified individually for every cell in the cross section.

TIRE GROUND-OUT CAPABILITYThe Large Deformation Element (LDE) developed and validated during the last

“The anisotropy of the tire is a direct consequence of the tire structure.”

F1 F1F1

F2 F2 F2

F2

F1

Ready for Print - 20.12.2013Ready for Print - 20.12.2013

5.1 5.15 5.2 5.25 5.3 5.35 5.4 5.45 5.5

4,000

3,000

2,000

1,000

0

-1,000

-2,000

-3,000

-4,000

-5,000

Longitudinal Force

Forc

e [N

]

Time [s]

5.1 5.15 5.2 5.25 5.3 5.35 5.4 5.45 5.5

30,000

25,000

20,000

15,000

10,000

5,000

0

Vertical Force

Forc

e [N

]

Time [s]

Measurement

Simulation

Measurement

Simulation

CDTire/Realtime

Robot control

SIMPACK Realtime

Motion cueing

Washout filter

Inverse kinematics

A1

A3

A2

A4

A5

A6


3RD PARTY PRODUCT | Axel Gallrein, Manfred Bäcker, Michael Burger, Fraunhofer-Institute ITWM Axel Gallrein, Manfred Bäcker, Michael Burger, Fraunhofer-Institute ITWM | 3RD PARTY PRODUCT

4 years has been transferred to the new model CDTire/3D. This element models the sidewall/belt/rim contacts for very large tire deformations (see Fig. 5). As an extension, we have added the ability to distinguish between the inside and outside part of the rim as well as added support for PAX-like run-flat systems.

PARAMETER IDENTIFICATIONTo determine the model parameters of CDTire in such a way that a specific tire is adapted, tire data sets and measurements must describe the physical properties of the tire. The stand-alone tool CDTirePI allows the engineer to identify the best parameter set by:

• importing measurement data from arbi-trary test labs

• setting up and executing the respective test scenarios, and

• automatically comparing test and simula-tion results using dedicated measures (e.g., rms, range, max, etc.)

In the first step, the cross section geometry is constructed. All material properties are interpreted relative to the reference geom-

etry. The import of constructional data is supported. Functional tire components like belt, carcass and cap plies, and tread are addressed as separate entities. Their specific geometric properties (such as cord angles) are fixed during this step.Next, certain tire (pre-strain) properties are adapted to align the cross-section geometry with measurements under inflation pres-sure. Both inner and outer contour and cross section mid-surface geometry can be used here.Quasi-static stiffness tests can now be used to adapt most elastic properties. In this phase, additional geometric information like footprint shapes and deformed contour measurements can be used to resolve some parameter ambi-guities introduced by pure 'spindle load' information.If no additional measurements are avail-able, an 80 % accurate solution can still be obtained via the structural capabilities of the model, if local information was used to parameterize it. This opens the door for tire types like large agricultural tires, where up to now, tire model usage has been difficult to

Fig. 5: Forced tire ground out simulated with CDTire/3D on Fraunhofer LDE test rig

Fig. 6: Driving simulator at Fraunhofer ITWM

calculate due to insufficient measurements. Usually, stationary measurements can be used to identify tread and friction proper-ties; cleat run tests asses the viscous proper-ties. For tires too big to perform cleat runs, drop tests and/or modal analysis can be substituted.

MODEL SCALABILITY OF CDTIRE/3DThe discretization of the material shell struc-ture is arbitrary and user selectable for side

wall and belt geom-etry in circumferential as well as in lateral direction.In addition to dis-cretization scalability,

users can speed up CDTire/3D by exchang-ing its material side wall model with an analytical discrete membrane model. If a CDTire/3D parameter set is available, the model parameters for the analytical sidewall model can be derived semi-automatically by using CDTirePI. With the analytical sidewall formulation, the simulation speed can be nearly doubled, while the accuracy is af-fected only insignificantly for very large tire deformations. One can switch from one for-mulation to another easily by setting a flag

in the parameter-file. Down-scaling is there-fore straightforward with this support; and the up-scaling of legacy (CDT40) parameter sets only requires little additional effort.

CDTIRE/REALTIMEFor comfort and durability applications in hard real-time systems such as driv-ing simulators (see Fig. 6), the model CDTire/Realtime was developed. While it features many of the same modeling techniques— such as the contact and ana-lytical sidewall model of CDTire/3D — a fur-ther downscaling of the lateral discretization was necessary to achieve hard real-time. To-gether with very efficient memory storage, access programming in pure C/C++ and an implicit Newmark integration scheme with the choice of deterministic settings or traditional step size and iteration control, it enables simulation scenarios with extreme requirements for computational perfor-mance — such as hard real-time. An offline version of CDTire/Realtime called CDTire/HPS (High Performance Solver) is available in SIMPACK 9.4.Meanwhile Fraunhofer ITWM and SIMPACK AG have integrated CDTire/Realime together with SIMPACK Realtime on the Fraunhofer

Fig. 7: Concept of Fraunhofer driving simulator RODOS

ITWM driving simulator RODOS (see publi-cation Burger, M., Baecker, M., Gallrein, A., Kleer, M.: "Integration eines detaillierten, flexiblen Reifenmodells in den Fraunhofer Fahrsimulator", VDI-Berichte 2211, 14. Inter-nationale VDI-Tagung Reifen-Fahrwerk-Fahr-bahn, 2013; further see article "SIMPACK Realtime", SIMPACK News, July 2013). Fig. 7 shows the overall integration concept of the driving simulator. The vehicle model and the tire models are running on a Con-current iHawk 12 core real time system, where 3 cores were used for the vehicle model itself; 4 cores for the tires (one core for each tire) and 1 core for the co-simula-tion administrative layer. This administra-tive layer also organizes and coordinates the data transfer to the driving simulator, namely the chassis acceleration and orien-tation as an input to the robot control (via dSpace CAN bus communication) and the steering wheel angle, as well as brake and throttle pedal positions given by the opera-tor as input to the vehicle model. Fig. 7 also shows a recorded actual maneuver with a sample rate of 1 millisecond consisting of a straight run over a series of 18 obstacles (heights varying from 15 to 25 millimeters), followed by a curve.

“Users can speed up CDTire/3D by exchanging its material side wall

model with an analytical discrete membrane model.”


Enveloping model with elliptical cams

Rigid ring (6 DOF)

Sidewall stiffness & damping

Effective road plane

Cleat

Rim

Residual stiffness & damping

Effective road surface

Contact patch dimensions

2000

1000

0

-1000

2000

0

-2000

2

0

-2

0.2

0.1

0

200

100

0

50

0

MeasurementSimulation

�𝘍𝘻

�N�

�𝘍𝘹

�N�

��rad/s�

time frequency

Cleat: H = 1; L = 5 cm𝘍𝘻0 = 4_kN; V = 39_km/h

Body input force of front left shock absorber

Body input force of front right shock absorber

Test

Simulationtime

time

forc

efo

rce

�S𝘍𝘻𝘍𝘻

�N/�Hz�

��

��

�S𝘍𝘹𝘍𝘹

�N/�Hz�

��

��

�S��

�rad �Hz�

��

��

Slip model


Antoine Schmeitz, TNO | 3RD PARTY PRODUCT3RD PARTY PRODUCT | Antoine Schmeitz, TNO

TNO’s MF-Tyre / MF-Swift and the Delft-Tyre Toolchain

Fig. 1: Schematic view of the model structure

BRIEF HISTORY OF THE MODELAfter it was first published in 1987 and 1989, Pacejka’s Magic Formula model quickly gained broad acceptance as a highly accurate model for describing measured steady-state forces and moments occur-ring under various slip conditions. Later extensions to the basic Magic Formula have been made to include some of the transient behavior of the tire, e.g., by introducing relaxation lengths. Today, the latest version is capable of dealing with combinations of brake slip, sideslip, camber, turn slip, infla-tion pressure and transient responses up to

Fig. 2: Typical cleat test validation result; 205/60 R15 passenger car tire

Fig. 3: Body input forces for vehicle driving over rough road; 19 inch run flat tires [2]

approximately 8 Hz. TNO’s first implementa-tion of the model (MF-Tyre) dates back to 1996.The need for more accurate vehicle dynamic simulations (also on uneven roads), includ-ing control systems like ABS and ESC, emerged in the 1990s. These applica-tions require the model to be valid for higher frequencies (> 30 Hz) and for short wavelengths (> 0.2 m), because many low frequency aspects of tire behavior are velocity-independent and can better be

expressed in terms of wavelength. This was the motivation for developing the Magic Formula-based Short Wavelength Interme-diate Frequency Tyre model (MF-Swift).In the first decade of this century, the em-

phasis of model develop-ment changed to impact harshness, ride comfort and road load (durability) applications. This resulted in the addition of a 3D

enveloping model and the inclusion and support of advanced road models for rough roads.

The development of MF-Swift has been carried out by TNO in close cooperation with Delft University, and later Eindhoven University. Throughout this time, the auto-motive and tire industries were continuously involved and supported the developments.

MODEL OVERVIEWA schematic view of the model is shown in Fig 1. The important elements of the MF-Swift tyre model are:

• Magic Formula• Contact patch slip model• Rigid ring• Obstacle enveloping model

• Magic FormulaThe well-known Magic Formula model of Pacejka can describe the steady-state tire slip forces and moments with high ac-curacy. Basically, this semi-empirical model consists of a set of equations that are pa-rameterized by fitting these to steady-state slip measurements.

• Contact patch slip modelA tire does not respond instantaneously to changes in slip; a certain distance must be traveled before the steady-state levels of forces and moments are reached. The tire relaxation behavior is caused mainly by the flexibility of the tire structure. This flex-ibility is included by elastically suspending the tire contact patch (via the rigid ring) to the wheel rim. For short wavelengths, the finite length of the contact patch must also be considered. This is accomplished by using the contact patch slip model that accounts for the contact patch transients.

Tire models for vehicle dynamic analysis must accurately represent tire behavior, be computation-ally fast, reliable and robust. In addition, when applying tire models to virtual vehicle develop-ment, it is essential that the required model parameters are well and efficiently identified from common experiments. Consequently, not only a tire model is required, but also a full methodology. For this reason, TNO developed the Delft-Tyre Toolchain. Toolchain and the tire models MF-Tyre/MF-Swift consist of a portfolio of products and services to comply with all tire modeling needs.

“The automotive and tire industry were continuously

involved and supported the developments.”

• Rigid ringIt appears that when considering a maximum frequency of approximately 60–100 Hz, the deformations of the tire belt can be neglected. Consequently, the tire belt is modeled as a rigid body/ring which is elastically suspended with respect to the rim. Residual springs are introduced

between the ring and contact patch to en-sure that the overall stiffness of the tire (and relaxation lengths) is correct.

• Obstacle enveloping modelWhen rolling over short obstacles and rough roads, tire geometry and elasticity give rise to the nonlinear behavior of the tire forces and to changes in the effective rolling radius. To incorporate this enveloping behavior, the concept of the effective road surface is used. This effective road consists of plane height, slope, curvature and bank-ing. An obstacle-enveloping model, consist-ing of elliptical cams that touch the actual road surface, is used to generate it. Finally, the single point rigid ring model contacts this road surface and the slip forces from the Magic Formula act on it.

This model structure and the software implementation allow the user to select the level of complexity. For handling analyses, it is generally sufficient to use only the Magic Formula, whereas for road load simulations, the full complexity is required.

USAGE AND RANGE OF APPLICATION• All kinds of vehicle steering and han-

dling simulations, e.g., ISO tests such as



3RD PARTY PRODUCT | Antoine Schmeitz, TNO Antoine Schmeitz, TNO | 3RD PARTY PRODUCT

Fig. 6: TNO Tyre Test Trailer, wet testing

Fig. 5: TNO Tyre Test Trailer, winter testing

Fig. 4: MF-Tool parameter identification software

steady-state cornering, lane changes, J-turn, braking, sine with dwell, Fishhook, µ-split, low µ, rollover, parking effort, etc.

• Vehicle behavior on rough roads: impact harshness; ride comfort analyses; road load calculations for durability analyses

important, as the quality of the process as a whole affects the final simulation accuracy, and the costs involved are relatively high. Consequently, it is required to have well-defined test procedures that can be carried out in common test facilities and to have tools available for identifying the tire model parameters. For this reason, TNO has developed the parameter identi-fication software MF-Tool (see Fig. 4), and offers services for parameter identification.The unique structure of MF-Swift has special advantages here, as the model consists of a number of relatively independent elements: the parameter identification process can be split up into a number of consecutive, small optimization problems in which a subset of parameters is uniquely determined. With MF-Tool, non-expert users can generate MF-Tyre and full MF-Swift tire property files within 10 and 60 minutes respectively.TNO is well experienced in conducting tire measurements. For the development of

MF-Tyre/MF-Swift, several unique tire test facilities were developed. In the most recent development, the TNO Tyre Test Trailer (Fig. 5 and 6) was adapted to perform winter testing under severe conditions.After the procedure and required tests to

obtain the model parameters, under-standing of the tire model is important for successful us-age. This is why

TNO has always been very open about its models and offers training and consulting services.

SUMMARY AND OUTLOOKTNO Delft-Tyre Toolchain is a portfolio of products and services that include the tire models MF-Tyre/MF-Swift, the parameter identification software MF-Tool and mea-surement, parameter identification, training and consulting services. MF-Tyre/MF-Swift is a versatile model that can be used for many applications. The complexity of the model can be selected for each simulation.

TNO is continuously developing its Tool-chain. Current research is focused on model-ing tire behavior under various operating conditions such as ice, snow, temperature, velocity, etc. In addition, more real-time ap-plications will be realized in the near future, e.g., for Hardware-in-the-Loop and driving simulator applications. Further improvement of the model for misuse applications is under investigation.

INFORMATION AND CONTACTFor an extensive description of the model, further references and many validation re-sults, please refer to reference [1].For information about tire model imple-mentation, please contact SIMPACK AG; for other questions, please find our contact information at www.delft-tyre.com.

REFERENCES[1] Pacejka, H.B., Tyre and Vehicle Dynamics, 3rdEdition, Butterworth-Heinemann, Oxford, 2011.[2] Schmeitz, A., Versteden, W., Eguchi, T., Road Load Simulation using the MF-Swift Tire and OpenCRGRoad Model, SAE Technical Paper 2011-01-0190, 2011.

• Simulations in which control systems are involved that can excite or be influenced by tire dynamics, e.g., ABS, ESC, etc.

• Analysis of driveline vibrations• Analysis of shimmy vibrations; typically in

the range of 10–25 Hz

• Lap time and fuel consump-tion simulations

• Passenger car, truck, motor-cycle, raceing, and aircraft tires

• Delft-Tyre products are used by 70 % of the automotive OEM.

MODEL VALIDATIONDuring the development of MF-Tyre/MF-Swift, extensive effort has been spent on model validation. Dedicated tire test rigs have been developed to investigate tire behavior and check model capabilities. Some examples of the validation tests used are: force and moment tests, dynamic braking, dynamic steering, cleat tests, axle height oscillations, stiffness tests, etc. In addition, instrumented ve-hicle tests have been conducted in cooperation with vehicle OEM to demonstrate the ap-

plicability of the model for applications in vehicle development. An example of a typi-cal cleat test is shown in Fig. 2; an example of a road load simulation is shown in Fig. 3.

TOOLS AND SERVICESIn general, tire testing and parameter identification effort for tire models are very

“Current research is focused on modeling tire behavior under various

operating conditions such as ice, snow, temperature, velocity, etc..”


Michael Gipser, Gerald Hofmann, cosin scientific software | 3RD PARTY PRODUCT3RD PARTY PRODUCT | Michael Gipser, Gerald Hofmann, cosin scientific software

➡ 500 to 2500 DOFs, assigned to a ring of belt segments:

translation along x/ y / z rotation about longitudinal axis bending with 3 to 9 shape functions

➡ Belt segments coupled to each other and to rim by a large set of nonlinear spring/damper elements, reflecting the tire's structural stiffness properties

hysteresis by dry friction

Ffriction ddyn

dradial

cfriction cdyn

belt node

dynamic stiffening by Maxwell elements

force/deflection progressivity run-on-flat bottoming

by nonlinear radial springs

rim

Ffriction

belt node

c radial

progression

d radial

c dyn

F friction d dyn

ddyn

cdyn

cfriction

belt node

cradial

progression


FTire: High-End Tire Model for Vehicle Simulation in SIMPACK

IDEAS DRIVING FTIRE DEVELOPMENTA modern tire simulation model is expected to be a virtual reproduction of the true tire, covering all of its vehicle dynamics-relevant aspects and their respective cross-correla-tion, and providing reliable insight into the dynamic behavior of the tire.Such an accurate tire model has to take into account many different kinds of external or internal excitations, for example:

• short-waved road irregularities (Fig. 1)• sharp-edged and/or high obstacles with

small extensions like cleats and stones• location- and time-dependent road fric-

tion properties and water film depth• high-frequency rim oscillations induced

by active suspension control systems• variable inflation pressure• variable tire and road surface temperature• actual tread wear state• tire imbalance, non-uniformity, run-out

and other imperfections

• rim flexibility• road surface flexibility and/or plasticity

Due to the nonlinear and high-frequency nature of most of the tire dynamics aspects, a physics-based model appears to be the only option. The intrinsic benefit of physics-based models is that there is no need to find and ‘build in’ the tire behavior for every single combination of tire input signals, tire states and operating conditions. Rather, as with the real tire, the behavior is a consequence of fewer but more reliable physical principles. The art of tire modeling is no longer the Si-syphean task of implementing ever more ar-tificial influencing factors, trying to take into account every new observed phenomenon. Rather, it comes down to deciding which physical effects may be neglected or simpli-fied under certain conditions. The remaining

• impulse and energy conservation• structure distortion under different kinds

of static loads• relationships between local and global

stiffness• symmetry and defined of linearized mass

and stiffness matrix, etc.

FTire arranges the physics-based tire de-scription into subsystems, well-defined sys-tem boundaries and interface signals which have clear physical meaning. This allows easy activation and deactivation of certain tire model extensions, without replacing the data-file or changing the model interfacing. To complete this 'model scope on demand' approach, two more aspects have to be mentioned:

• FTire is scalable with respect to timely and spatial resolution of structural dis-cretization and road surface sensing. This is achieved by strictly decoupling physical data and numerical settings. Whatever internal time step, segment number or tread block number is chosen, the pre-scribed static, modal and steady-state properties used to determine internal physical parameters will be precisely matched by FTire. This is achieved in a fully automated way, during the so-called data pre-processing phase, which is re-peated if numerical settings are changed.

• Several of the subsystem models were introduced to extend FTire’s scope of ap-plication, e.g., road surface description,

soft-soil model and rim flexibility model, which may be replaced with user-written versions using standardized C-code interfaces.

By the means listed above, the tire model variant actually used is tailored to the ap-plication, saving computing time without the need to maintain different tire models or tire model data files.

The most important development aspect of FTire is the design and implementation of feasible, reliable, and affordable methods to determine the parameters of the physical submodels.After this, the inherent numerical complex-ity of some of the models requires the de-velopment of optimized ODE and PDE solv-ers, running in co-simulation with the MBS, FEM, or system dynamics solver. Today, it is expected that tire models run in real-time, for HiL and driving simulator applications. FTire does so, after some internal tuning, but with the original core model.

CORE MECHANICAL MODELFTire’s core model is a special, MBS-like ar-rangement of nonlinear elasticity, dissipative and inertia elements which replace the real tire structure (Fig. 2a/b). Selection of these elements and their parameterization is such that belt distortion under a wide range of relevant external loads on flat or non-flat surface matches that of the real tire in a de-tailed way. Lower-order eigenfrequencies, mode shapes (including bending modes), and modal damping values are matched sufficiently well at the same time. Stiffness parameters and internal belt normal forces are influenced by actual inflation pressure (Fig. 2c).The related parameterization takes respec-tive measurements of the global tire stiff-ness under different conditions, as well as eigenfrequencies and related mode shapes, for use in a parameter identification (PI) pro-

Fig. 1: Tire structure distortion on Belgian block road

simplified model still observes all related physical principles. To a certain extent, and in clear contrast to pure mathematical ap-proximations, such models are able to ex-trapolate conditions not completely covered by a respective experiment. For example, consider the real tire’s main structure. This structure is composed of a ‘nearly’ axisymmetric placement of carcass,

belt, and bead, ar-ranged in layers and embedded into dif-ferent types of rubber compound as matrix

material. In FTire, SIMPACK’s high-end tire model, this structure is replaced by a closed chain of small nonlinearly flexible bodies. Despite this obvious simplification, the fol-lowing are automatically observed:

• well-known relationships between differ-ent modes and mode-shapes, and their dependency on load and rolling speed

“A modern tire simulation model is expected to be a virtual reproduction of the true tire...”

Fig. 2b: Structure model: radial force elements

Fig. 2a: Structure model: belt segments

The tire is undoubtedly both the most important and the most complex vehicle suspension component.With this in mind, tire simulation today requires more than the mathematical approximation of a hand full of steady-state, deflection- and slip-based force/moment characteristics on ideally flat or con-stantly curved surfaces. And it requires more than

sensing the road below the tire at a single point, or an ‘equivalent volume’ ap-proach, which condenses the complex

geometrical road surface shape into one value.

Neglecting this would result in over-simplified tire dynamics and rough road

influence. This is what 'classical' tire models have done since the earliest vehicle dynamics

simulations.Instead, the increasing complexity of state-

of-the-art vehicle models requires a versatile, robust and multi-purpose tire model that can be

used not only in 'classical' handling application scenarios, but also with highly dynamic road excita-tions, misuse scenarios, and any component test rig application, simultaneously.



internal stiff roads

pressure forces:

Fpressure = F (p) • n

stiffness:c = c (p)

user defined

stiff road

interface

internal soil model

user defined

soil model

interface

FTire core

stiff rim

internal flexible rim model

user defined

flex. rim

model interface

internal MBS susp. model

3rd Party MBS solver

3rd Party system

simulator

CTI

CTI

> 0.800.70 .. 0.800.60 .. 0.700.50 .. 0.600.40 .. 0.500.30 .. 0.400.20 .. 0.300.10 .. 0.20< 0.01

t= 0.152ss= 0.84hgrid-line dist. 20mm

ground pressure [MPa]


Fig. 2c: Inflation pressure influence

Fig. 5: Tread wear model activated

Fig. 4: FTire: subsystems, system borders, interfaces, plug-ins

Fig. 3: Tread model

cess. Alternatively, comparable static load cases or modal analyses of a FEM tire model may serve as 'virtual measurements'. In or-der to compare modal data, FTire provides linearized system matrices at any arbitrary operating point.The second component of the core model is the tread description, comprised of an arbitrary number (typically between 3.000 and 10.000) of massless 'contact ele-ments' (Fig. 3). These contact sensors, with extensions in the range of millimeters, are attached to the structure model elements and potentially come into contact with the road. If so, radial deflection (and thus local ground pressure), local sliding velocity, indi-vidual temperature, and local road contact tangent planes are used to compute sensor-individual normal and friction forces. These forces constitute the distributed external load of the structure model.By setting the unloaded lengths of the contact sensors in an appropriate way, it is possible to take into account grooves, lugs, and other tread patterns (Fig. 3).FTire’s road evaluation supports all quasi-industry-standard road data formats, and in addition, has a simple interface for user-defined road models. The complete decoupling of road evaluation from SIM-PACK allows SIMPACK to be connected to all FTire-supported road models, without any extra provisions within SIMPACK.

SUBSYSTEMSNext to the core model, FTire optionally pro-vides several extensions, most of which are

implemented in internal, switchable subsystems (Fig. 4):

• Extra contact elements for tire misuse simulations, like belt-to-rim contact (bot-toming), sidewall-to-curb contact, and rim-to-curb contact• Several modifications of the core mode parameters, to take into account differ-ent types of tire imperfec-tions such as imbalance, non-uniformity, tread gauge variations, conicity, ply-steer

and more• A thermal model, predicting

the filling gas temperature as well as the tread surface temperature field.

It is driven by heat generated in all dis-sipative and friction elements of the me-chanical core model, as well as by cooling through convec-tion, radiation and a transfer of heat into the environment. In turn, the temperatures influence both inflation pressure (and by this indirectly

structural stiffness), and tread friction characteristics

• A tread wear prediction model, driven by local friction power and tread tempera-ture (Fig. 5)

• A flexible and visco-elastic rim model, which can be replaced by a user-provided model

• A soft soil model, based on the Bekker-Wong soil equation [7] which can be replaced by a user-provided terra-me-chanical soil model (Fig. 6)

• A fluid-dynamics-based filling gas vibra-tion model, driven mainly by time- and location-dependent variations of the tire cross section, to predict excitation and influence of 'cavity modes'

PARAMETERIZATIONIn FTire's parameterization procedure, a clear distinction is made between data

used internally in the model equations ('pre-processed' data), and data to be supplied by the user ('basic' data). Basic data are obtained

by standard laboratory measurements, or by equivalent simulations with an FEM model, if available.

“A clear distinction is made between data used internally in the model equations, and

data to be supplied by the user.”

These standard measurements are selected to be as inexpensive, repeatable, signifi-cant, and as reliable as possible. Observing this, a standardized measurement proce-dure — which was recently defined by a working group of German car manufactur-ers — has been recommended.Comprehensive software (FTire/fit, Fig. 7) is available to automate the parameterization and validation process (Fig. 8a/b), based upon these measurements. However, such methods require significant training and expertise in tire dynamics. Certain test labo-ratories provide 'turn-key' FTire data files.

INTERFACING AND NUMERICSFTire is run in co-simulation, thus completely decoupling the integration of its huge number of internal state variables from SIMPACK’s DAE solver. As with all other supported simulation environments, the in-terface between SIMPACK and FTire is real-ized by an easy to apply, but comprehensive application programming interface. This API, called CTI (cosin tire interface), handles:

• exchange of system signals between ve-hicle and tire model

• loading of tire and road data files• specification of operating conditions• requests for extra output

• control of FTire’s animation• provision of TYDEX/STI output• provision of key tire data, as required by

the calling vehicle model

• selection of user-defined submodels• specification of FTire’s ‘speed mode’, a

collection of settings to influence FTire’s speed of computation (up to real-time


0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

400

200

0

-200

-400

longitudial force

Directly Measured Data

size & geometry, mass vertical stiffnes....

Traction / Braking

tread stiffness...

In-Plane Statics

belt in-plane and lateral bending stiffness, belt longitudinal stiffness...

In-Plane Cleat Tests

belt extensibility, in-plane damping, more tread rubber properties...

Handling small slip values

belt-out-of-plane bending stiffness...

Out-Of-Plane Statics

belt lateral and torsional stiffness...

Out-Of-Plane Cleat Tests

out-of-plane damping, belt out-of-plane flexibility kinematics...

Friction Characteristics large slip values

sliding friction coefficients

Wheel Load 2.3 kN Wheel Load 3.5 kN Wheel Load 4.7 kN

Wheel Load 2.3 kN Wheel Load 3.5 kN Wheel Load 4.7 kN

longitudial force longitudial force

vertical force vertical force vertical force

longitudial force longitudial force longitudial force

vertical force vertical force vertical force

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40time [s] time [s] time [s]

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40time [s] time [s] time [s]

[N]

0.00 0.02 0.04 0.06 0.08 0.10

time [s] time [s]

0.00 0.02 0.04 0.06 0.08 0.10time [s]

0.00 0.02 0.04 0.06 0.08 0.10

0.00 0.02 0.04 0.06 0.08 0.10time [s] time [s]

0.00 0.02 0.04 0.06 0.08 0.10time [s]

0.00 0.02 0.04 0.06 0.08 0.10

[N] [N]

[N] [N] [N]

[N] [N] [N]

[N] [N] [N]

600

400

200

0

-200

-400

-600

800

600

400

200

0

-200

-400

-600

800

600

400

200

0

-200

-400

-600

1500

1000

500

0

-500

-1000

1500

1000

500

0

-500

-1000

1500

1000

500

0

-500

-1000

-1500

4000

3500

3000

2500

2000

1500

1000

5000

4500

4000

3500

3000

2500

2000

6000

5500

5000

4500

4000

3500

3000

3600

3400

3200

3000

2800

2600

2400

2200

2000

4400

4200

4000

3800

3600

3400

3200

5600

5400

5200

5000

4800

4600

4400

> 0.800.70 .. 0.800.60 .. 0.700.50 .. 0.600.40 .. 0.500.30 .. 0.400.20 .. 0.300.10 .. 0.20< 0.01

t = 0.091ss = 1.63hgrid-line dist. 20mm

ground pressure [MPa]



capability for simultaneous computation of all tires of a vehicle)

and much more. The use of these capa-bilities is under sole control of the calling solver. CTI and FTire are packed into a single dynamic library (cti.dll in Windows systems and libcti.so in Linux systems), without any extra dependencies.

Fig. 6: FTire on soft soil model: pressure distribution and road surface deformation

Fig. 8b: Cleat test validation, v = 40 km/h (blue = measurement, red = FTire)Fig. 7: Sequential parameter identification, using direct data, static properties, steady-state measurements, and dynamic cleat tests for in-plane and out-of-plane excitation

Fig. 8a: Cleat test validation, v = 5 km/h (blue = measurement, red = FTire)

FTire’s specialized ODE/PDE solver updates tens of thousands of state variables (includ-ing 3D displacements, temperature, friction and wear state of all contact elements) once per internal time step. The duration of this time step can be cho-sen, but is typically around 0.2 ms.

The integrator deals with potentially extremely large non-linear structure defor-mations, locally unstable friction character-istics, and high numerical stiffness of the structure’s equations of motion.The integrator takes full advantage of FTire’s numerical properties, like the nearly axisym-metric tire structure and clear discrimination between stiff and non-stiff system com-ponents. It guarantees certain static and modal properties of the discretized model, independent of actual numerical settings like internal step-size and mesh resolution.Although using a constant internal step-size is preferable, CTI and FTire can be connect-ed to any step-size- and order-controlling external solver. However, FTire’s perfor-mance is best if the solver step size does not change too often.

CONCLUSIONNearly 15 years of intense development, preceded by 15 years of experience in tire modeling, have made FTire the most

comprehensive and frequently used physi-cal tire model today. Providing the complete tool chain to create,

analyze, and process data for both tire and road surface properties, it is available

in all important vehicle simulation environ-ments. FTire is used by several dozen OEMs, tire manufacturers, tier-1-suppliers, and research institutes world-wide and can be applied to almost all types of ground vehicle and aircraft tires. It has become the first choice for ride comfort, handling, durability, and mobility applications.

REFERENCES[1] FTire documentation and more material: www.cosin.eu[2] Gipser, M.: FTire — the Tire Simulation Model for all Applications Related to Vehicle Dynamics. Vehicle System Dynamics, Vol. 45:1, pp. 217–225, 2007[3] Gipser, M.: RGR Road Models for FTire. Proc. SAE World Congress 08M-54, Detroit 2008[4] Gipser, M.: The FTire Tire Model. In: Pacejka,H. B.:

Tyre and Vehicle Dynamics (3rd ed.), pp. 582–586. SAE International 2012[5] Gipser, M.: Pneumatic Tire Models: the Detailed Mechanical Approach. In: Road and Off-Road Vehicle System Dynamics Handbook. Mastinu, Plöchl (eds), CRC Press, to appear in November 2013[6] Wong, J.Y.: Terramechanics and Off-Road Vehicle Engineering (2nd ed.). Butterworth-Heinemann, Oxford 2010

“A modern tire simulation model is expected to be a virtual reproduction of the true tire...”


CE

PEE

SEE

F x

F x

F tan

F tan

F wheel

Fy

Fy

Hill’s Equation

The muscles’ behavior in the model is described through Hill’s 1938 empirical investigated equation.

In Hill’s Equation, the hyperbolic connec-tion between muscle force and contrac-tion velocity is mathematically formulated.

�𝐹�� + �� 𝑣�� ― �� = ―�𝐹��|𝑣�� =� + ��

forceF(v) / FMAX

100

50

150

-0.5 0 0.5 1.0

contraction velocity v / vMAX


Matthias Nusser, Veit Wank, Eberhard Karls University of Tübingen; | CUSTOMER APPLICATION Valentin Keppler, Biomotion Solutions GbR

CUSTOMER APPLICATION | Matthias Nusser, Veit Wank, Eberhard Karls University of Tübingen; Valentin Keppler, Biomotion Solutions GbR

and the activation pattern for each of the 16 muscles were recorded (Fig. 4). A planar motion analysis was carried out with high speed videos and markers at-tached to the subjects joints (Fig. 2). Video analysis has been done with sub pixel accu-racy by automatic pattern tracking using the software TraXXol. To enhance the accuracy of the measured motion data, large white markers have been placed on the subject who was additionally wearing a black elastic full body suit to maximize the contrast. The collected pedaling kinematic data was used to calculate an average motion pattern of a ped-al cycle. To enhance the accuracy of pedal torque data, we captured the crank angle with a high speed camera and motion analysis, too. For reliable analysis, the data (pedal forces, crank-torque, both perspectives of the camera, as well as EMG-signals) was time synchronized.

MUSCLE ACTIVATION DURING SIMULATIONAs a first simulation approach, the neces-sary activation patterns for the simulation were generated from the EMG data. The data was always synchronized with the current angle of the crank. The EMG-signal of each muscle was collected and averaged throughout up to 20 cycles. The identified strength of the signals, collected from the highest isometric contraction, was used for the standardization. The activation patterns gained through EMG-analysis were used to control the simulation. This has been done by co-simulation between SIMPACK and Simulink®. The input for the controller given by the SIMPACK model was the crank angle. The Simulink model calculated the new stimula-tion vector for each leg (in total 16 activa-tion signals) by use of a look-up table.

Musculoskeletal Model of Bicycle Pedaling

Fig. 1: Musculoskeletal model of the lower limb

The project for "modeling and simulating cycling based on a musculoskeletal model" was carried out as a thesis at the Institute of Sports Science, Sports Biomechanics and Exercise Science at the University of Tübin-gen, Department of Biomechanics, under the supervi-sion of Professor Dr. Veit Wank and in cooperation with the company Biomotion Solutions GbR. The following text provides a short summary of the project and its results.

THE MECHANICAL MODELThe foundation of our simulation was a musculoskeletal model of the lower extremities, which was formerly used at the University of Tübingen to calculate the pressure distribution within the hip joint [1]. The human body has been modeled by 23 rigid bodies. Data for mass and inertia of the body segments have been taken amongst other sources from NASA publications [2].Most of the essential, anatomical data on the musculoskeletal model of the lower limbs was based on work by Scott L. Delp [3, 4]. The complex model for the lower extremities consisting of more than 40 mus-

cles per leg was reduced to the simulation of these muscles that could be measured reliably. Therefore sixteen muscles (8 at each leg), generating moments for the cyclic movement of the legs during bicycling were adopted into the model. The mechanical properties of each muscle were implemented in the model as con-tractile element (CE) parallel elastic element (PEE) and a serial elastic element (SEE). The mathematical formulas described by A.V. Hill [5] were used as a base for the force development of the muscle force element in

the model. As the muscles are acting along a com-plex path along the bones of the lower

limbs, each muscle has been realized as a muscle force controller which calculated the muscle force and by one or more point to point force elements to actuate the bones with the muscle forces calculated by the controller. Some of the muscles are acting on a muscle path which calls for muscle wrapping. Therefore eight moved markers have been implemented at each femur. As some muscles in the leg can “lift off” at some of the contact points if the legs are in a stretched position, these moved markers allow the muscle to lift off the femur when knees are straight and to wrap around the femur and patella if the knees are bent. The bicycle model consists of 6 bodies (crank, 2 pedals, fly wheel, chassis, and saddle). A torque element acting on the flywheel enabled the simulation of different

The objective of this project was to simulate the pedaling movement of cycling based on a modified musculoskeletal model, and to compare the results with the actual movement. The different activation patterns of the leg muscles necessary for the simulation were collected, prior to simulation, with an ergometer. The goal was to gain insights from the results about the functionality and capacities of our

model so further modifications and developments could be put into action.From a practical viewpoint, a model like this can help prevent injuries, for example, in easily-injured areas like the patella. Especially in the area of rehabilitation, bicycle ergometers are widely used and integrated into therapy. The creation of a high capacity simulation model, which will enable the diagnosis of muscle strength and strains, can improve therapy quality. Such a model can also improve competitive and recreational sports, for example, through optimal saddle positioning.

Fig. 3: Muscle model with contractile element (CE) and elastic structures (PEE, SEE)

Fig. 2: Biomechanical motion analysis using high-speed video data

Fig. 4: Pedal forces measured by strain gauge system

Fig. 5: Tangential and radial pedal forces acting at the crank

“The collected pedaling kinematics data was used to calculate an average

motion pattern of a pedal cycle.”

EXPERIMENTAL SETUPFor further comparisons between simulated and real pedaling movement at the Karl-sruhe Institute for Technology (KIT), data was collected in order to identify important physiological and biomechanical parameters of the performance of the bicycle rider. We used a scientific ergometer build by SRM (Schoberer Rad Messtechnik, www.srm.de) which was equipped with a 2D pedal force measurement device (Powertec®) which al-lowed for capturing the acting forces during pedaling in tangential and radial direction. The Powertec system measures the pedal forces by two sensors which determine the magnetic field variations (Hall-Effect) as a result of the displacement in respect to a magnet [6]. At different pedaling frequen-cies and power settings, together with the forces acting on the pedals, the total power

brake torques to simulate, e.g., pedaling against constant power or velocity depen-dent brake torque. The feet of the human body model were connected to the pedals by force elements to simulate click-in pedals which were used during the measurements. The human body model was connected to the saddle by a bushing-like force element.

F wheel

As the EMG-signals for both legs differed slightly, we decided to symmetrize the artificial EMG-Data so, we substituted the measured EMG-signals for the left leg by the 180 degree shifted signals from the right leg.

RESULTSThree simulations were done in total, each with different adjustments. In the first simulation, the power was kept at a con-stant level; in the second simulation, the effective crank-torque was kept constant.

The third simulation was performed using a speed-dependent damp-ing for generating the crank-torque. A clear

match was identified between measured and simulated crank speed. The minimum values of crank velocity occur in the upper and lower turning point; the maximum val-ues occur on horizontal pedal position.

“The simulation has been done by co-simulation between SIMPACK and SIMULINK.”


M.rectus femoris R [mV]

M.tensor facia latae R [mV]

0.00 100.00 200.00 300.00

0.400

0.300

0.200

0.100

0.000

0.3500.3000.2500.2000.1500.1000.0500.000

0.3000.2500.2000.1500.1000.0500.000

0.3000.2500.2000.1500.1000.0500.000

0.6000.5000.4000.3000.2000.1000.000

0.400

0.300

0.200

0.100

0.000

0.800

0.600

0.400

0.200

0.000

0.5000.4000.3000.2000.1000.000

M.soleus R [mV]

M.gastrocnemius medialis R [mV]

M.biceps femoris R [mV]

M.vastus lateralis R [mV]

M.gluteaus max.R [mV]

400W–70U Measurement

crank angle [*]

0.00 100.00 200.00 300.00

480.00

460.00

440.00

420.00

400.00

cran

k sp

eed

re [*

/s]

0.00 100.00 200.00 300.00

460.00

450.00

440.00

430.00

420.00

410.00

cran

k sp

eed

re [*

/s]

crank angle [*]

400W–70U Simulation damping: proprtional to omega

460.00

440.00

420.00

400.00

cran

k sp

eed

re [*

/s]

crank angle [*]

400W–70U Simulation const.P

0.00 100.00 200.00 300.00

M.tibialis anterior R [mV]


Steve Mulski, SIMPACK AG; Jochen Harms, elb |sim|engineering | SOFTWARECUSTOMER APPLICATION | Matthias Nusser, Veit Wank, Eberhard Karls University of Tübingen; Valentin Keppler, Biomotion Solutions GbR

460.00

440.00

420.00

cran

k sp

eed

re [*

/s]

crank angle [*]

0.00 100.00 200.00 300.00

400W–70U Simulation const.M

Looking at the animation or the joint angle velocities of the human body model revealed that the approach to activate the muscles by measured data is not sufficient to get a smooth and effective pedaling motion. This is based on the fact that our model contains only feed forward control so

Fig. 8: Crank velocity of an average cycle in measurement (top) and results of simulations with different crank reaction torques

Fig. 6: Muscle activation patterns of pedaling measured with EMG

Fig. 7: Muscle activation patterns were measured through electromyography (EMG)

Sciences 126.843 (1938): 136–195. [6] Stapelfeldt, B., et al; "Development and evaluation of a new bicycle instrument for mea-surements of pedal forces and power output in cycling"; International journal of sports medicine 28.04 (2007): 326–332.

far. The lack of feedback partially leads to subop-timal stimulation timing. But the results of this study are very promis-ing, and the model will be extended with a closed loop control for muscle stimulation.

SUMMARY AND VIEWThe simulation of the cyclic leg movement while cycling, based on EMG-measured muscle activation patterns, shows a correspondence with the collected data. In the first approxima-tion, it can be proved that the model is valid. But it is obvious that modification and fine tuning of the muscle activation routines is necessary. To further fine tune the model, it may be necessary to pay attention to the smaller muscles of the leg. Specifically it can be assumed that imple-mentation of a feedback

loop for muscle stimulation control might enhance the quality of the results.The generated model together with the SIMPACK model is a good basis for expan-sion and modification of the existing model, to possibly create more realistic results in future simulations.

BIBLIOGRAPHY[1] Prochel, Anton; "Erstellung eines komplexen Muskel-Skelett-Modells zur Berechnung der Druckbelastung in Gelenken bei vorwärtsdyna-misch simulierten Bewegungsformen" (2009)[2] Chandler, R.F. Clauser, und C.E. MCConville; "Investigation of inertial properties of the human body"; AMRL Technical Report, NASA Wright-Patterson Air Force Base, 74, 1975[3] Delp, Scott L., and J. Peter Loan, "A graphics-based software system to develop and analyze models of musculoskeletal structures", Comput-ers in biology and medicine 25.1 (1995): 21–34[4] Delp, Scott L., et al; "An interactive graphics-based model of the lower extremity to study orthopaedic surgical procedures"; Biomedical Engineering, IEEE Transactions on 37.8 (1990): 757–767.[5] Hill, A. V.; "The heat of shortening and the dynamic constants of muscle"; Proceedings of the Royal Society of London. Series B, Biological

Load Calculations for Wind TurbinesSIMPACK scripts are now available for easily defining, running and analyzing the thousands of load cases required for proper dimensioning of components and fulfilling certification requirements for wind turbines. The load calculation process can be used with any SIMPACK wind turbine model, i.e., from simple models used for initial concept studies and certification purposes to detailed models used for component analysis, optimization and extreme event investi-gations. Because these scripts are freely available to all customers and are fully documented, users can easily further extend the existing functionality to include any company specific enhance-ments and/or future requirements.

HISTORY OF SIMPACK IN THE WIND INDUSTRYInitially used as a high-end tool for investi-gating potential resonances and maximum loading of individual components, such as within gearboxes, SIMPACK has seen its way from being a simulation tool used in a reserved area of analyses, to becoming software applicable to all aspects of wind turbine simulation. Over the years, many new features have been developed to assist users in generating and investigating wind turbine behavior. For example, the Rotorblade Generator module used for easily creating flexible blades out of standard cross-sectional blade data; inter-faces to state-of-the-art aerodynamic codes; and interfaces to standard wind turbine controllers (to name just a few). Added to this, the SIMPACK solver technology, world

Fig. 1: Load case definition window

renowned for speed, accuracy and stability, along with countless other features, such as the Gear Pair module, initially developed for Formula 1 engine simulations, and inter-faces to Simulink, SIMPACK has now gained a significant presence within the wind turbine sector. More recently, several wind

turbine companies have invested signifi-cant effort to develop their own in-house solutions for carrying

out and processing multiple parallel load calculations with SIMPACK. After initial calculations, component models can then be easily exchanged with higher fidelity models for more specific analyses (Fig. 2). In order to assist companies in setting up these standard processes, fully documented SIM-PACK scripts have now been developed and are available for any SIMPACK user. These scripts represent another major milestone

Fig. 2: Example of wind turbine models with different component models

in the history of SIMPACK wind turbine simulation.

PROCESS REQUIREMENTSResulting from collaboration with com-mercial partners within the wind sector, the SIMPACK scripts build upon years of experience gained from company specific in-house software solutions for load calcula-tions. The main objectives of the load calcu-lation scripts can be summed up as follows:

• An intuitive window based GUI (Graphical User Interface) for entering simulation runs

• Use with any SIMPACK wind turbine model, including any level of model detail

• Easy entry of load cases with automatic generation of wind fields

• Additional ability to vary model specific parameters

• Automatic job allocation on multiple processors

• Easy review of results for plausibility• Fast and parallel statistical analysis• Saving and retrieval of simulated models,

load configurations and results• Open QSA (Qt Script for Applications)

scripts for user specific enhancements

The final requirement was of particular im-portance to our partners. Open scripts can be enhanced to fulfill any company specific requirements, or even future requirements, thus ensuring the longevity of the loads process.

LOADS PROCESS: PRE, SOLVING, POSTOnce the Load Calculation Script has been started, the loads process is as follows:

“...users can easily further extend the existing functionality to include any company specific enhancements...”


$B_DT2 rigid

1

1$B_DT1 rigid2

2

$B_NCL rigid

$B_YAW rigid

$B_TWR_connect rigid

1

2101

1

1

a

43

22

11

430

a

0

0

11

1

1

1

1

DLC11_WS0009_WD000_UF008_HE0200_TS0006

time [s] time [s]

time [s] time [s]

time [s] time [s]

0 100 200 300 400 500 600 700

rota

tiona

l spe

ed [r

pm]

angl

e [d

eg]

velo

city

[m/s

]

torq

ue [k

Nm

]an

gle

[deg

]

torq

ue [k

Nm

]fo

rce

[kN

][-]

[-]

0 100 200 300 400 500 600 700

0 100 200 300 400 500 600 700

0 100 200 300 400 500 600 700

0 100 200 300 400 500 600 700

0 100 200 300 400 500 600 700

16

14

12

10

8

6

4

Hub total wind velocity Hub xy wind direction Rotor thrust

Pitch angle 1 Tower bottom MyPitch angle 2 Pitch angle 3

Generator speed Shaft brake statusGenerator torque Generator status Safety system

0.00.00.00.00.00.00.0 0.00.0

-0.0

12001150110010501000950900850800

1.2

1.0

0.8

0.6

0.4

0.2

0.0

-0.2

70

60

50

40

30

20

10

x 103

x 103

0.9

0.8

0.7

0.6

0.5

0.4

0.3

20151050

-5-10-15-20-25

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

x 103

time [s]

Win

d ve

locit

y [m

/s]

25

20

15

10

5

0

x 103

Gen

erat

or to

rque

[Nm

]

x 103

50

40

30

20

10

0

Wind velocityGenerator torque

0.0 0.2 0.4 0.6 0.8 1.0 1.2

time [s]

Defle

ctio

n x

[m]

Pitc

h al

gle

[deg

]

x 103

6

5

4

3

2

1

0

-1

25

20

15

10

5

0

-5

Blade deflTower deflPitch angle

0.0 0.2 0.4 0.6 0.8 1.0 1.2

Cycles

5.0e+0064.5e+0064.0e+0063.5e+0063.0e+0062.5e+0062.0e+0061.5e+0061.0e+0065.0e+0050.0e+000

5.0e+0064.5e+0064.0e+0063.5e+0063.0e+0062.5e+0062.0e+0061.5e+0061.0e+0065.0e+0050.0e+000

Load

0.0e+000 1.0e+007 2.0e+007 3.0e+007 4.0e+007 5.0e+007

6.0e+007 7.0e+007

8.0e+007 7.0e+007 6.0e+007 5.0e+007 4.0e+007 3.0e+007 2.0e+007 1.0e+0070.0e+000

Mean


SOFTWARE | Steve Mulski, SIMPACK AG; Jochen Harms, elb |sim|engineering Steve Mulski, SIMPACK AG; Jochen Harms, elb |sim|engineering | SOFTWARE

Fig. 5: Ultimate loads Fig. 6: Markov results

1. Loads simulation folder A folder is chosen in which all models, time domain results and statistical analyses will be stored.

2. Simulation Model A selected simulation model is copied to the loads simulation folder. This model is then used as the basis model for all following simulations.

3. Check Plot configurationA user predefined SIMPACK Plot File (spf) is selected in which important characteristic simulation results have been plotted, e.g., wind speed and direction, generator power, hub and tower loads, pitch angle, etc. After each simulation, this file is used as a tem-plate for creating the check plots, in png format, for easy review (Fig. 3).

4. Execution of the initialization runsInitial conditions are generated for each simulation. This minimizes initial transients which require simulation time to settle down (Fig. 4). State Sets, which include all model states, e.g., pitch angles, rotor speed, bending of blades and tower, are generated for each wind speed and wind direction. Each defined simulation run uses the cor-responding State Set for the model’s initial condition.

5. DLC Definition Generation of all variations (Fig. 1) required for each Design Load Case (DLC).The entered variations include:

• Simulation parameters (directory, DLC name, simulation time, etc.)

• Wind conditions (mean speed, direction, type, turbulence seed)

• Controller configuration• Aerodynamic configuration• User defined parameters

6. Wind field generation All required wind fields for the defined DLCs are automatically generated using TurbSim for turbulent wind fields, and IECWind for transient wind fields.

7. Simulation The user can set the number of CPU cores to be used. Job allocation is automatic. A Check Plot is generated as soon as an indi-vidual job is complete.

8. Statistical analysis Once the simulation jobs are complete, several statistical analyses can be executed. Lists of result files are defined by the user and then allocated to a particular analysis method. Safety and Weibull distribution factors are entered for ex-treme predictions and fatigue analysis. The individual output channels of the result file are predefined within SIMPACK Post.Several scripts are available for the statistical analysis:

• Ultimate loads (Fig. 5)• Generation of input data,

power production time se-ries, for load extrapolation tools

• Rainflow count algorithm; calculation and output of Markov-Matrices, Load Spectrum and Damage Equivalent Load (DEL) calculation (Fig. 6)

• Load Duration Distribution (LDD); calcula-tion of the Load Spectrum with multi-dimensional output, up to five channels

• Load Revolution Distribution (LRD)• ASCII export of time series (e.g., for use

with fatigue tools)

9. Saving and selecting of load configurations All GUI entries are stored within text based ASCII files. These files can be loaded and used as a basis for defining further con-figurations. Since the files are text based, the load process can also be run as a batch process from the command line.

MODEL DESCRIPTIONAn example wind turbine model is available with the scripts (Fig. 7). This model is com-parable in detail to the NREL 5 MW model. In addition to demonstrating the Loads Calculation process within SIMPACK, this model serves as a good starting point for wind turbine dynamists new to SIMPACK. The model consists of several submodels (foundation, tower, nacelle and rotor). The rotor is further broken down into submodels of the hub and blades. All submodels are fully parameterized and can also be easily exchanged with corresponding models in order to vary the level of detail. The tower is a parameterized SIMBEAM model, and the rotor blades have been generated using the SIMPACK Rotorblade Generator. Aerody-namics are integrated using AeroDyn v13. A DLL (Dynamic Link Library), which complies

with the Garrad Hassan standard, is used for the control.

PACKAGE DESCRIPTIONIncluded within the SIMPACK documenta-tion, 9.6 and above, is the demonstration wind turbine model, load calculation scripts and documentation. The scripts consist of solver scripts for the load calculation process and Post scripts for the statistical analysis. The scripts have been created us-ing a modular structure with an individual script for each statistical analysis. The scripts are also internally commented in order to assist users wishing to extend their func-tionality. The docu-mentation contains a step-by-step guide on how to set-up, configure and run load calculations; and how to carry out statistical analyses with the scripts.

FURTHER POSSIBILITIESDetailed models of drivetrains, pitch and yaw systems are easily substituted for the available submodels. Detailed FE (Finite Ele-ment) models of gearbox housings, planet carriers, bedplates, etc., may also be used. Advanced rotorblades containing coupled twist/bend terms are easily generated and implemented within the model. Component foundation models for on- and off-shore wind turbines can also be included. In es-sence, any SIMPACK wind turbine model, regardless of the level of detail, can be used with the load calculation scripts. Since simulation time depends upon the detail contained within the model, users may easily generate model variants which fit the requirements of a particular analysis and have optimal simulation times.The DLC (Design Load Case) examples do not cover all requirements within the current guidelines but, with the included parameter variation functionality, all necessary load cases can be readily configured.

The available script for statistical analysis provides a solid basis for generating all design relevant loads. All scripts, including those for statistical analysis, can be easily enhanced to fit within any design process. In order to enable user specific interfaces and/or user routines to comply with the load calculation process, some scripts may need to be adapted.

LICENSINGFor the design and certification of wind turbines, often several thousand simulation runs are required. Computers with many

multi-core proces-sors are commonly used within the wind industry in order to speed-up

these load calculations. A specific SIMPACK solver license package is available which enables multiple solver jobs to be carried out simultaneously. This package is only available for simulating load calculations of wind turbines and some restrictions do apply. More information about this solver package may be acquired from your local SIMPACK distributor.

CONCLUSIONFully documented and commented scripts are now available for carrying out load calculations with SIMPACK wind turbine models. These scripts can essentially be used with any wind turbine model, regard-less of the level of modeling detail, for any load case. All SIMPACK users have access to these scripts which can be readily enhanced to suite any company specific process. A specific solver license package enables mul-tiple simultaneous job runs.With the addition of Load Calculation functionality within SIMPACK, wind turbine users can now readily carry out analyses of initial concept designs through to detailed component investigations of final designs all within one software tool, and all with one database of interchangeable components.

Fig. 3: Check Plot

Fig. 4: Initialization runFig. 7: Example wind turbine model with nacelle submodel (3D and 2D view)

“These scripts can essentially be used with any wind turbine model, regardless

of the level of modeling detail,...”


bogie frame secondary suspensionbogie bolster

wheelset

bogie pin

primary suspension axle box

housing

axle box


Gernoth Götz, Bombardier Transportation GmbH | CUSTOMER APPLICATION CUSTOMER APPLICATION | Gernoth Götz, Bombardier Transportation GmbH

from On-Track Tests for Model ValidationTHE DYNOTRAIN PROJECTIn 2007, the European Commission (EC) ini- tialized the Seventh Framework Programme (FP7) for research and technological develop-ment. DynoTRAIN was one of three projects in the TrioTRAIN cluster (Total Regulatory Acceptance Interoperable Network), started in June 2009 (scheduled project dura-tion: 48 months, extended to 52 months) with funding by the EC under FP7 (Grant

Agreement no. 234079) with a budget of € 3.3 million. DynoTRAIN included the fol-lowing objectives: closing of open points in the high speed (HS) and conventional rail (CR) technical specifications for interoper-ability (TSIs) related to vehicle dynamics; harmonize European and national stan-dards on railway dynamics; and reduce costs of certification and development of an inno-vative certification process using computer

simulations. The research work was divided into seven work packages (WP):

WP1: Measurement of track geometry quality and virtual certification

WP2: Track geometry qualityWP3: Contact geometryWP4: Track loading limits related to

network accessWP5: Model building and validationWP6: Virtual certification of modified

vehicles and vehicles running in other conditions

WP7: Regulatory acceptance

The DynoTRAIN project team is a unique international consortium co-ordinated by UNIFE, the European Rail Industry. In WP5, 14 partners from industry (Bombardier, Siemens, Alstom, CAF, AnsaldoBreda), universities, public research institutes (TUB, KTH, POLIMI, IFSTTAR, CEIT, RSSB) and transport companies (DB, Trenitalia, SNCF)

collaborate. For more information, see www.triotrain.eu.

DYNOTRAIN MEASUREMENT CAMPAIGNIn October 2010, the German transport company Deutsche Bahn (DB) compiled a 325 m long measurement train. Over a four-week period, on-track measurements were carried out with slightly different train

configurations over a 5000 km distance through Germany, France, Switzerland and Italy. The measurement train consisted of 13 different rail vehicles:

• One electric locomotive DB series 120.1 (normal operation mode, with measuring equipment)

• One inter-city passenger coach series Bim 547.5 (normal operation mode, empty, with measuring equipment)

• Two flat freight wagons with Y25 bogies for containers and swap bodies series Sgns 691 (one wagon empty, one wagon laden, wheelset load of 22.5 t, both with measuring equipment)

• Freight wagon unit consisting of two flat wagons with stakes series Laas (one wagon empty, one wagon laden, wheel-set load of 20.0 t, both with measuring equipment)

Fig. 1b: MD36 bogie model of a Bim 547.5 coach

The rail vehicle approval process today is costly and time-consuming. "Virtual testing" using multi-body simulations is one approach to reduce these costs. During the European research project DynoTRAIN, one of the main activities in work package 5 was the comparison of simulations and measurements from on-track tests to develop the process and criteria for model validation. The on-track measurements were carried out with a measurement train in Germany, France, Switzerland and Italy. This ar-

Fig. 1a: MD36 bogie of a Bim 547.5 coach

ticle describes some selected simula-tions with the Bim 547.5 coach using the simulation tool SIMPACK, which have been carried out by Bombardier Trans-portation.

INTRODUCTIONFor the past two decades, Europe has in-creased its cross-border transport of freight and passengers. The European Union is aware that this makes cross-border rail transport a viable alternative to road trans-port. Historically, every country in Europe de-veloped its own railway system with different national rules for test-ing the acceptability of railway vehicles’ running characteristics. Today, it is necessary for cross-border rail transport to be interoperable with various railway systems. The European approach to interoperability led to two European Commission (EC) Council Directives: 96/48/EC on July 23, 1996 on the interoperability

of trans-European high-speed rail systems and 2001/16/EC on March 19, 2001 on the interoperability of the conventional rail systems. At both of these councils, processes to determine standard rail vehicle approval were established. Today, testing vehicle running characteristics is a costly and time-consuming process since the vehicle certi-fication against European standards (EN), like EN 14363 [1], requires multiple field tests. Unexpected environmental or other

boundary conditions influence the results so that field tests have to be repeated several times in order to cover the possible

range of circumstances, increasing costs and duration of vehicle approval. One approach to reduce this effort could be "virtual test-ing" using numerical multi-body simulations (MBS). For the application of this methodol-ogy, the validation of the simulation model is essential.

“Today, it is necessary for cross-border rail transport to be

interoperable with various railway systems.”

Comparison of Simulation and Measurements


-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6

25

20

15

10

5

0

-5

-10

-15

-20

-25

Torq

ue [k

Nm

]

Angle [º]

Measurement 1Measurement 2Simulation (initial model)Simulation (adjusted model)

1.028 1.027 1.026 1.025 1.024

80

70

60

50

40

30

20

Whe

el lo

ad Q

[kN

]

Distance [m] x105

Simulation initial modelSimulation adjusted modelMeasurements


Gernoth Götz, Bombardier Transportation GmbH | CUSTOMER APPLICATION CUSTOMER APPLICATION | Gernoth Götz, Bombardier Transportation GmbH

Fig. 3: Bogie rotation test results in comparison with simulation using initial vehicle model and model after adjustment

all four test zones according to EN 14363 [1] were defined to carry out the comparisons with the on-track tests. Furthermore, differ-ent groups of stationary tests were defined to compare their results with simulations. The sequence of the simulation activity was as follows: initially, the simulation model based on technical drawings and data sheets was only compared with on-track tests in straight sections with a check on the wheel loads (to adjust the car body mass properties), but without comparison to stationary tests. To analyze the effect of measured boundary conditions, the simu-lation results, based on different com-binations of measured and standard boundary conditions, were compared to the on-track measurements. In the next step, the simulation model was compared to defined stationary test groups. If pos-sible, the simulation model was improved to reduce the difference between simulation and measurement of stationary tests. Sub-sequently the same on-track comparisons had been carried out again. The purpose of the simulation activity was to highlight the information about the need for sta-tionary tests and track parameters (wheel profile, rail profile and track irregularity). Fig. 2 shows the comparison of the simulat-ed and measured Q forces at the first wheel-

set based on one exemplary track part from the Swiss line Biasca – Göschenen (curve radius R = 295 m, velocity v = 74 km/h, superelevation u = 150 mm). To make the calculation as realistic as possible, bound-ary conditions (measured wheel and rail profiles, measured track geometry and track irregularity data) were used. Additionally, in Table 1 some selected calculated wheel-rail forces and some selected acceleration RMS values are shown. The comparisons with stationary tests and other adjustments led

to a final model. The selected results under the same boundary conditions were also included in Fig. 2 and Table 1.

By using the final adjusted model, the qua-sistatic differences between simulation and measurements in the Q-forces were slightly reduced. This was mainly dependent on the adjustment of the height of the car body center of gravity. Fig. 3 shows exemplary comparisons with stationary tests on the results of the bogie rotation test. The main difference between the initial simulated and measured torque hysteresis curves can be observed for rota-tion angles between ±4 and ±4.5 degrees. In the initial simulation model, the charac-teristic of the secondary longitudinal bump stop between the bogie frame and bolster includes a free play. This characteristic was

estimated and represented an uncertain model parameter. The approach used to improve the simulation model was a reduc-tion of this free play to zero and a new bump stop characteristic. The comparison of the improved model demonstrates very good agreement between simulation and measurement.

CONCLUSIONThe complete analysis of the simulation results of all vehicles and the proposed process and criteria for model validation has been published in the final WP5 deliverable and at conferences IAVSD 2013 in Qingdao, China [3] and Bogie '13 in Budapest, Hun-gary [4]. Based on the proposed validation methodology in WP5, the Bim 547.5 vehicle model fulfils the validation limits and can therefore be regarded as validated. The ve-hicle model is ready to use for virtual vehicle approval. The following conclusions can be drawn regarding the comparisons between simulation and measurements:

• Measured track irregularities as well as measured rail and wheel profiles improve the accuracy of simulation results com-pared to on-track measurements.

• Stationary tests can be used for model im-provements if there are uncertain vehicle parameters. Due to a good data basis for the Bim 547.5 coach, the vehicle model improvement by comparisons with sta-tionary tests was only marginal; the main improvement was achieved by compari-sons with on-track test measurements.

REFERENCES[1] EN 14363 Railway Applications — Testing for the Acceptance of Running Characteristics of Railway Vehicles — Testing of Running Behaviour and Stationary Tests, CEN, Brussels, 2005.[2] Schelle, H., Hecht, M.: Testing for the Ac-ceptance of Running Characteristics: Influence of Boundary Conditions on the Simulation, ZEV Rail, August 2013[3] Polach, O., Böttcher, A.: A new Approach to Define Criteria for Rail Vehicle Model Validation. 23rd International Symposium on Dynamics of Vehicles on Roads and Tracks, Qingdao, China, August 19–23, 2013[4] Polach, O., Böttcher, A., Vannucci, D., Sima, J., Schelle, H., Chollet, H., Götz, G., Garcia Prada, M., Nicklisch, D., Mazzola, L., Berg, M., Osman, M.: Validation of Multi-body Models for Simulations in Authorisation of Rail Vehicles. 9th International Conference on Railway Bogies and Running Gears, Budapest, Hungary, 9–12 September, 2013

“To analyze the effect of measured boundary conditions,

the simulation results, were compared to the on-track measurements.”

• One DB RAILab coach (DB Rolling Analy-sis and Inspection Laboratory for the measurement of track geometry, track irregularities and rail profiles)

• One measuring coach with equipment for data recording of locomotive, pas-senger and freight coaches

• Six brake coaches

During the test campaign, the vehicle dynamics of the locomotive, the pas-senger and the freight coaches were measured using the following mea-surement equipment:

• Instrumented wheelsets to measure the rail-wheel contact forces Y (lateral force), Q (vertical force) and Tx (longitudi-nal force)

• Acceleration sensors to measure the vertical and lateral accelerations of axle boxes, bogie frames and car body

• Displacement sensors to measure the relative displacements in the primary and secondary suspension.

At each vehicle, around 50 measuring channels were recorded with a maximum sampling rate of 1200 Hz depending on the recorded quantity. The

Table 1: Comparison of quasistatic wheel-rail forces and car body accelerations with initial and adjusted simulation model (line Biasca – Göschenen, curve radius 278 m, velocity 74 km/h, superelevation 150 mm)

Fig. 2: Wheel loads first wheelset, line Biasca – Göschenen, curve radius 278 m, velocity 74 km/h, superelevation 150 mm

recorded data provided the data basis for a comparison of measurement and simulation in DynoTRAIN WP5.

BIM 547.5 SIMULATION MODELThe Bim 547.5 passenger coach is part of the standard UIC-X wagon series that DB has used since 1952 in D-trains and inter-city trains. Bombardier Transporta-

tion simulated this passenger coach with SIMPACK using technical drawings and data sheets (Fig. 1). The initial model consisted of one car body and two bogie series Minden Deutz 36 (MD36). All mass bodies were modeled as rigid bodies. The bogie con-tained conventional wheelsets, axle guides for wheelset guidance, flexicoil springs for primary and secondary suspension and dampers in different spatial directions. In comparison to freight wagons, the passen-ger coaches have for ride comfort reasons a secondary suspension level (MD36: bogie bolster with flexicoil springs). Friction forces occur between the bogie bolster and the car body and also in the secondary spring support. The vehicle has a maximum velo-city of 200 km/h, a wheelset base distance of 2.5 m, a bogie base distance of 19.0 m and a wheel diameter of 0.95 m.

DYNOTRAIN WP5 SIMULATIONSOne of the main activities in WP5 was the comparison of simulation and measure-ments from on-track tests to develop the process and criteria for model validation. The importance of boundary conditions (measured wheel and rail profiles, mea-sured track irregularties) [2] and model improvements by stationary tests were also investigated. 17 track sections representing


-0.6 -0.4 -0.2 0 0.2 0.4 0.6Velocity (m/s)

1.5

1.0

0.5

0

-0.5

-1.0

-1.5

Forc

e (k

N)

TheoreticalExperimental

𝑦(𝑘)

𝑘�

𝑘�


Liu Wei, Vehicle Engineering, Qingdao University | CUSTOMER APPLICATIONCUSTOMER APPLICATION | Liu Wei, Vehicle Engineering, Qingdao University

Table 1: Parameters of phenomenological model

sion systems with MR dampers. The full-car model of the light bus, with detailed suspen-sion and steering system, was carefully built in SIMPACK. The MR damper was designed and manufactured as an important semi-actuator. Based on the work developed by Spencer et al. in 1996, a mathematical model of an MR damper was established. From experimental data, the behavior of the proposed model was fitted to real device behavior. With the neural network controller and mathematical model of the MR damper, a closed loop control system for semi-active suspension was established.

(1)

where the damping force is governed by

(2)

Solving (1) results in

(3)

where, 𝐹� is the damping force of the MR damper, and the values of unknown factors (𝐴, 𝛽, 𝛾, 𝑛, 𝑐0, 𝑐1, 𝛼, 𝑘0, 𝑘1) require testing to determine.In order to get the parameters of the phe-nomenological model, an MR damper test rig was established. Fig. 2 shows the experi-mental configuration for damping force con-trol of the MR damper. The MR damper is excited by an ampli-tude of ±20.0 mm and frequency 2.0 Hz using a hydraulic exciter. The damping force of the MR damper is measured by a force

sensor, and piston velocity is measured by a displacement sensor. Control signal is generated from a computer system with a D/A board, and this signal is fed back to the current amplifier and applied to the MR damper.The results of the parameter identification process using the optimization algorithm are shown in Table 1.Fig. 3 shows the comparison of theoretical and experimental results. It can be seen that the theoretical predictions are in close agreement with the experimental results. This indicates that the mathematical model can be used to describe the characteristics of the prototype MR damper.

DESIGN OF SEMI-ACTIVE CONTROLLERA semi-active controller aims to reduce the motion of the vehicle body and the vertical vibration of unsprung masses under differ-

ent operating condi-tions. Generally speak-ing, ride comfort and handling stability are conflicting objectives.

How to impose different control strategies under different operating conditions is the main topic to be discussed next.

“...the theoretical predictions are in close agreement with the experimental results.”

MR DAMPER MODELA prototype MR damper with mixed mode operation was designed and manufactured in order to perform as a semi-active suspen-sion system in the vehicle. The nonlinear force-velocity and force-displacement responses of the MR damper have been described by many mathematical models, such as the Bingham model and nonlinear hysteretic bi-viscous model, etc. [9, 10]. One of the most popular is the Bouc-Wen hysteresis models, originally developed by Bouc in 1967 and then modified by Wen in 1976 [11, 12]. Later on, Spencer proposed a phenomenological model for MR dampers, a more accurate design based on the Bouc-Wen model [13], shown in Fig. 1. This model predicts the force-displacement behavior of the damper accurately, and it expresses force-velocity behavior that more closely re-sembles the experimental data. The damp-ing force of the MR damper can be derived as follows:According to the nonlinear force-velocity and force-displacement responses of the MR damper, the mathematical model of MR damper is expressed as equation (1, 2, 3):

Fig. 3: Theoretical and experimental data of MR damper

Fig. 4: PID neural network control algorithm block diagram

neural network

PID controller semi-active suspension𝑟(𝑘) �(𝑘)

+

𝐾�𝐾�𝐾�

��(𝑘)/��

―X

INTRODUCTIONIn the process of automotive suspension design, ride comfort and handling stabil-ity are two conflicting considerations. In normal driving conditions, the automobile needs "soft" suspensions to achieve better ride comfort; in turning conditions, "hard" suspensions are required to reduce the heel-ing angle of the automotive body, for better handling stability and turn portability. This means that, under different driving condi-tions, automobiles require different suspen-sion features. In the past few decades, a great deal of research was conducted on semi-active suspension design [1, 2, 3]. However, while most of these semi-active

Fig. 1: Phenomenological model of MR damper

Kinematics Analysis and Design Optimization of Semi-Active Suspension

for a Light Bus

Fig. 2: Experimental configuration of MR damper

suspensions contribute to vehicle ride com-fort, they do little to improve automobile handling stability. In recent years, the controller design for sys-tems with complex nonlinear characteristics has become challenging, and applications of neural networks in the automotive industry have made many remarkable achievements [4, 5, 6]. The approximate neural network

model be can substituted for the actual model in many applications, especially con-trol system design. With appropriate off-line training, the neural network controller can help semi-active suspensions reach a good balance between ride comfort and handling stability [7, 8].The goal of this article is to propose a neural network controller for semi-active suspen-

For this article, first written as a paper for the SAE Interna-tional 2011 Vehicle Battery Summit, a Magneto-Rheo-logical (MR) fluid semi-active suspen-

sion system was tested on a commercial vehicle — a domestic light bus — to de-termine the performance improvements compared to passive suspensions. MR fluid is a material that responds to an applied magnetic field with a signifi-cant change in its rheological behavior. When the magnetic field is applied, the properties of such a fluid can change from a free-flowing, low-viscosity fluid to a near solid. This change in proper-

ties takes place in a few milliseconds and is fully reversible. A quarter sus-pension test rig was built to test out the nonlinear performance of the MR damper. Based on a large number of experimental data, a phenomenological model of an MR damper based on the Bouc-Wen hysteresis model was adopted to predict both the force-displacement behavior and the complex non-linear force-velocity re-sponse. Comparison with experimental results indicated that the mathematical model could effectively portray the be-havior of a typical MR damper and was adequate for control design and analy-sis. In order to accurately simulate the

performance of the commercial vehicle, a detailed multi-body dynamic model of the light bus with four semi-active sus-pensions was established, and an actual vehicle handling and stability test was carried out to verify the correctness of the multi-body dynamic model. For the

purpose of develop-ing a semi-active controller, the theory of neural network control is adopted here to identify and

control the semi-active suspension sys-tems. The primary goal of this article is to create an effective, reliable and safe semi-active suspension that provides ride comfort as well as handling stabil-ity of the commercial vehicle.

“In recent years, the controller design for systems with complex

nonlinear characteristics has become challenging.”

Bouc-Wen

𝐹�𝑐�

𝑦 𝑥

𝑐�

D/Aboard

current controlexcitation

control

hydraulic unit

force sensor

A/D board

acquisition system

test bench

displacement sensor

current amplifier


0 1 2 3 4 5 6Time (sec)

5

4

3

2

1

0

-1

-2

-3

-4

-5

Roll

acce

lera

tion

(m/s

2 )

0 1 2 3 4 5 6Time (sec)

5

4

3

2

1

0

-1

-2

-3

-4

-5

Pitc

h ac

cele

ratio

n (m

/s2 )

0 1 2 3 4 5 6Time (sec)

5

4

3

2

1

0

-1

-2

-3

-4

-5

Verti

cal a

ccel

erat

ion

(m/s

2 )



flects the rate of change of the deviation; 𝐴3 = 1 means the biasing logic. 𝐵1 = 𝑒(𝑎) is the deviation of the ideal and the actual ac-celeration of unsprung mass; 𝐴2 = 𝖽𝑒(𝑎)/𝖽𝑡 reflects the rate of change of the deviation; 𝐴3 = 1 means the biasing logic. The output of this node matches the degree of an input to the corresponding membership functions in the PID set.

(5)

(6)

where, 𝑜1, � is the membership grade of a PID set.On the hidden layer, each node represents an "AND" operator, and is a fixed node whose output is the product of the entire incoming signals. The inputs and outputs of hidden layers can be expressed as:

(7)

(8)

where, �1 is the valve value of the hidden layer. 𝑤1� is the connected weight between the input layer and the hidden layer. 𝑔(𝑛𝑒𝑡��) is Sigmoid type active function, that can be expressed as.

(9)

where �1 is the parameter representing the biasing value. �0 is used for regulating the shape of the Sigmoid function.The input of the output layer is:

(10)

where 𝑤�� is the connection weights of the hidden layer neuron i and the output layer neuron 𝑙; �� is the valve value of the output layer neuron. The outputs of the output layer correspond to the three parameters of the PID controller, 𝐾�, 𝐾� and 𝐾�. The basic principle of the learning algorithm is the gradient steepest descent method, that is, adjusting the weights of the network to minimize the network errors. Select the performance index function as follows:

Fig. 8: Block diagram of MATLAB and SIMPACK co-simulation

(11)

According to the gradient method, the cor-rection formula for the connection weights of output layer neurons is:

(12)

where η is the learning rate, η > 0. The con-nection weights of the output layer can be expressed as:

MATLAB® and Simulink®

current

ControllerMRF damper

values required by controller

vertical displacment of

damper pistons

forces of dampers reactionsSIMAT

(co-simulation interface via IPC)

sampling period: 0.001s

Server remote: offServer adress: 127.0.0.1

Server port: 20000

Auto start: off

��

�

Fig. 11: Roll acceleration of vehicle body (red = passive, black = semi-active)

Fig. 10: Pitch acceleration of vehicle body (red = passive, black = semi-active)

Fig. 9: Vertical acceleration of vehicle body(red = passive, black = semi-active)

The main purpose of a semi-active controller is to increase or decrease the damping coef-ficient in order to change the suspension damping force transient, which can reduce the vertical vibration of suspension and the heave, pitch and roll motion of the vehicle body. PID control is a more classic control algorithm, composed of proportional, inte-gral and differential links. When the vehicle yaw rate and instability are too large, the incremental PID control-ler can adjust the damping of suspension to produce a corrected yaw moment, and achieve the goal of vehicle body vibration suppression. However, there are couplings among heave, roll and pitch motions of vehicles, and it is very difficult to improve the ride comfort and handling stability of

an automobile simultaneously. Therefore, the neural network methodology is adopted here to identify the different driving condi-tions and adjust the proportional, integral and differential coefficients (𝐾�, 𝐾� and 𝐾�) of PID controllers. Finally, a hybrid damping

force control scheme based on PID and neural network con-trol methodology was adopted (Fig. 4). The input of this control algorithm, shown in

Fig. 4, is the difference between ideal value and actual value, and the output of this control algorithm is the added value of the controlled object. The incremental PID con-trol methodology is adopted in this article,

which provides the incremental of damping coefficient 𝛥𝑐 for semi-active suspensions. The PID control algorithm can be expressed as:

(4)

where, 𝑒(𝑘) is the deviation of the actual and target value and 𝑢(𝑘) is the output of the PID controller.Artificial neural network is developed within the biological neural network, and the basic unit of the biological neural network is the biological neuron, while the artificial neuron is the basic unit of artificial neural networks. Neural network controllers have been ex-pressed in many forms; a frequent represen-tation is a multilayer feed-forward network. In neural network representation, it can be easy to visualize and analyze the signal flow though the incremental PID controllers. A simple neural network control system prototype is shown in Fig. 5. It has two inputs (heeling angle and vertical accelera-tion of the vehicle), two input membership functions, and six outputs (proportional, integral and differential coefficients of PID controllers).Every node in the input layer, which is a membership function, is an adaptive node. There are six input nodes in this layer.𝐴1 = 𝑒(𝜑) is the deviation of the ideal and the actual heeling angle; 𝐴2 = 𝑒(𝜑)/𝖽𝑡 re-

Fig. 6: Dynamic multi-body model of suspensions Fig. 7: Dynamic multi-body model of vehicle

Fig. 5: Neural network control system prototype

“The main purpose of a semi-active controller is to increase or decrease the damping coefficient in order to change the suspension

damping force transient.”

output

𝐾�

𝐾�

𝐾�

𝐾'�

𝐾'�

𝐾'�

input 1heeling angle

input 2vertical

acceleration

input layer hidden layer output layer

�� AND

AND

AND

AND

AND

AND

AND

AND N

A1

A2

A3

B1

B2

B3

N

N

∑

��

��

�'��

�'��

�'��

AND


0 1 2 3 4 5 6Frequency (Hz)

3.5

3

2.5

2

1.5

1

0.5

0

Roll

acce

lera

tion

pow

er s

pect

ral d

ensit

y (ra

d/s2 )2 /H

z


3.5

3

2.5

2

1.5

1

0.5

0

Pitc

h ac

cele

ratio

n po

wer

spe

ctra

l den

sity

(rad/

s2 )2 /Hz


3.5

3

2.5

2

1.5

1

0.5

0

Verti

cal a

ccel

erat

ion

pow

er s

pect

ral d

ensit

y (m

/s2 )2 /H

z



signal transmission paths. In the first case, the physical model of the full car passes the values required by the neural network controller through the SIMAT module of SIMPACK. Then, the neural network con-troller analyzes the input signals and creates an output signal to control the intensity of currents. The magnetic intensity of the MR damper model is altered by the changes in input currents until lastly, the forces of the dampers’ reactions put off by the MR damper model are entered into the semi-active suspension of the full-vehicle model.In order to verify the accuracy of the neural network control strategy, the passive and semi-active suspension vehicle was simu-

Fig. 19: Time-domain data of front suspension vibration (red = passive, green = semi-active)

Fig. 21: PSD of front suspension vibration (red = passive, green = semi-active)

Fig. 20: Time-domain data of rear suspension vibration (red = passive, green = semi-active)

lated separately in a comparative analysis. In ride comfort simulation conditions, the light bus was assumed to be driving at 80 km/h driving on grade B road. In this situation, the vertical, pitch and roll acceleration of the vehicle body were tested to verify the effect of semi-active suspensions (see Fig. 9, 10 and 11, where the red line shows the time domain curve of passive suspension and the black line the time domain curve of semi-active suspension). The time domain signals were also used for frequency domain analysis, and the vertical acceleration power spectral density (PSD) curves, pitch acceleration PSD curves and roll acceleration PSD curves have been ob-

tained (see Fig. 12, 13 and 14, the red line showing the PSD curves of passive suspen-sion, and the black line the PSD curves of semi-active suspension). As shown in these figures, the curve amplitude of semi-active suspensions is much smaller than that of passive suspensions, which means the semi-active suspensions controlled by the neural network controller could be better able to reduce the vibration energy from the road. In handling stability simulation conditions, the snaking motion test condition was used to evaluate the handling performance of the light bus. The vehicle speed was assumed to be 50 km/h, and the heeling angle and yaw angular velocity of the vehicle body were tested to verify the effect of semi-active suspensions (see Fig. 15 and 16, the red line showing the time domain curve of passive suspension, and the black line the time domain curve of semi-active suspension). As shown in these figures, with the application of semi-active suspensions, the magnitude of heeling angle and yaw angular velocity of the vehicle body have been effectively controlled.

TEST RESULTSAn MR semi-active suspension test and con-trol system were set up and implemented on the light bus equipped with four MR controllable dampers. In the ride comfort

Fig. 22: PSD of rear suspension vibration (red = passive, green = semi-active)

0.00 10.00Time [s]

4.10

-4.10

Acce

lera

tion

of s

prun

g m

ass

(m/s

2 )

0.00 10.00Time [s]

3.80

-3.80

Acce

lera

tion

of s

prun

g m

ass

(m/s

2 )

0.00 50.00Frequency (Hz)

2.40

0.00Acce

lera

tion

pow

er s

pect

ral d

ensit

y [(m

/s2 )2 /H

z]

0.00 50.00Frequency (Hz)

2.30

0.00Acce

lera

tion

pow

er s

pect

ral d

ensit

y [(m

/s2 )2 /H

z](13)

According to the gradient method, the cor-rection formula of the connection weights of hidden layer neurons is:

(14)

The connection weights of the hidden layers can be expressed as:

(15)

The formula for the connection weights of output and hidden layer neurons 𝑙 and 𝑖 under the training of sample 𝑝 can be ex-pressed as:

(16)

(17)

AUTOMOBILE MODEL AND SIMULATIONThe light bus model used here should be a complete and relatively complex representa-tion of the actual vehicle. The major com-ponents of automotive virtual prototyping are discussed below. The virtual prototyping models of rear suspension established in SIMPACK are shown in Fig. 6. An element with an adjustable damping coefficient, also known as a semi-active damper, is an im-portant part of this model, and the damping force would be controlled by the semi-active suspension controller.Tires were modeled using the Magic For-mula tire force model which comes with SIMPACK. The inputs for the tire were obtained from the tire data provided by tire performance tests and were entered into the tire property file. The light bus model composed by drive, suspension and steering systems is shown in Fig. 7.To simulate the ride comfort and handling stability of the light bus, a neural network controller and virtual prototyping model of the automobile were established re-spectively in Simulink® and SIMPACK. By defining the data exchange interface between the integrated controller and the virtual prototyping model, the control of semi-active suspensions was achieved in the simulation environment. SIMPACK contains various kinds of random roads to be used in simulation tests. The interface standard provided by SIMPACK, SIMAT, makes the co-simulation with the Simulink package possible (Fig. 8). The four semi-active damp-ers with adjustable damping coefficients are controlled by Simulink with a neural network control module. The communica-tion interface between the software makes the connection between physical models of the dampers and controller possible. In the cooperating applications, there are three Fig. 18: Location of acceleration sensors

Fig. 17: Sensor and acquisition instrument

Fig. 16: Yaw angular velocity of snaking motion simulation

Fig. 15: Heeling angle of snaking motion simulation

Fig. 14: PSD curves of roll acceleration (red = passive, black = semi-active)

Fig. 13: PSD curves of pitch acceleration (red = passive, black = semi-active)

Fig. 12: PSD curves of vertical acceleration (red = passive, black = semi-active)

0 5 10 15 20 25Time [s]

4

3

2

1

0

-1

-2

-3

-4

Heel

ing

angl

e [d

eg]

0 5 10 15 20 25Time [s]

20

15

10

5

0

-5

-10

-15

-20

Yaw

ang

ular

vel

ocity

[deg

/s]



CUSTOMER APPLICATION | Liu Wei, Vehicle Engineering, Qingdao University

Modeling, Simulation and Dynamic Analyses of a Closed Single-Track Vehicle

The need for mobility and transpor-tation are becoming increasingly important. More energy-efficient trans-portation options — including electric vehicles — are being developed to meet the performance demands of today’s clients. The Monotracer is a comfortable single-track electric vehicle protected from weather and accidents by a closed cabin.This study focuses on the Monotracer. It will illustrate the fundamentals of driv-ing dynamics for this single-track vehi-cle, and reproduce them in a simulated model. At a later stage in the project, the pendulum and oscillating behavior

of the Monotracer at low speeds will be analyzed.

BACKGROUND AND KEY PROJECT CHALLENGES At its current stage of development, the Monotracer exhibits oscillating motions at low speeds on straight roads. The driver can feel these vibrations, particularly through the handlebars. The strength of the vibra-tion depends on the vehicle’s speed; the slower the Monotracer goes, the greater the amplitude of the pen-dulum motions. The strongest oscillations occur between 30 km/h and 50 km/h. At 60 km/h, the Monotracer stabilizes itself and drives straight without oscillating behavior.

Such oscillations are observed in typical motorcycles only at very high (> 250 km/h) or very low speeds (< 12 km/h).

GOALS AND OBJECTIVES In cooperation with the FHNW University of Applied Sciences and Arts Northwestern Switzerland, a simulation-capable model of the Monotracer was created and validated.

Vehicle parameters were also important, and had to be defined accordingly. Perfor-mance tests were also

necessary to define the model. With these, the driving dynamics simulations of the program could be optimized and adapted to the program. Through such processes, the influence of geometry on driving be-

“...a simulation-capable model of the Monotracer was created and validated.”

Fig. 1: Peraves AG Monotracer

test, test and control system included four acceleration sensors, A/D and D/A boards, a dSPACE Auto-Box control system with the control software, four MR dampers and four electronic current drivers, and a data acqui-sition instrument. The acceleration sensors were adopted to measure the accelerations at various points on the front and rear sus-pensions. The data acquisition instrument used in the ride comfort test is shown in Fig. 17. The location of acceleration sensors placement is shown in Fig. 18.When vehicle speed reaches 80 km/h on a random highway road, the vertical accel-

eration of front and rear sprung masses are shown in Figs. 19, 20, 21 and 22 (the red line showing the passive suspension, and the green line the semi-active suspension).The handling stability testing of the light bus with semi-active suspensions is shown in Fig. 23. In addition to the testing instru-ments used in the ride comfort experiment, a gyro was used to test the heeling angle and yaw angular velocity of the vehicle body. In the handling stability test, six stakes were laid out on the test site (30 m spacing), and the test bus passed through all stakes at a constant speed of 50 km/h (shown in

Fig. 23).The heeling angle and yaw angular velocity of the vehicle body is shown in Figs. 24 and 25 (the red line show-ing the passive suspension, and the green line the semi-active suspen-sion). As seen from these figures, the semi-active controller applied on MR dampers effectively reduces the roll and yaw motions of the vehicle body, and the effect of semi-active suspen-sion is reflected in the handling stabil-ity test.

CONCLUSIONThe ride comfort and handling stabili-ty results confirm that the semi-active controller based on a neural network can effectively suppress the vertical vibration of the suspension and the heave, pitch and roll motion of the vehicle body. The conflict between ride comfort and handling stability can be solved, and both aspects can be improved simultaneously.

REFERENCES[1] Alex, R. and Wang, P. Z., “New Resolu-tion of Fuzzy Regression Analysis,” Pro-ceedings of the IEEE International Confer-ence on Systems, Man and Cybernetics, vol. 2, pp. 2019–2021, 1998.

[2] Kazemian, H.B., “Developments of Fuzzy PID Controllers,” Expert Systems, vol. 22, no. 5, pp. 254–264, 2005.[3] Peng, B.B. and Huang, X.Q., “A Simulation Test Method for a Half Semi-active Vehicle Suspension Based on the Hierarchical Modeling Method,” 2006 IEEE International Conference on Vehicular Electronics and Safety, ICVES, pp. 63–67, 2006.[4] Cong, S. and Liang, Y., “PID-like Neural Net-work Nonlinear Adaptive Control for Uncertain Multivariable Motion Control Systems,” IEEE Transactions on Industrial Electronics, vol. 56, no. 10, pp. 3872–3879, 2009.[5] Hahm, D., Koh, H.-M., Ok, S.-Y., Park, W., Chung, C. and Park, K.-S., “Cost-effectiveness Evaluation of MR Damper System for Cable-stayed Bridges Under Earthquake Excitation,” Proceedings of the 3rd International Conference on Bridge Maintenance, Safety and Manage-ment — Bridge Maintenance, Safety, Manage-ment, Life-Cycle Performance and Cost, pp. 301–302, 2006.[6] Guo, A.X., Cui, L.L. and Li, H., “Structural Control of Seismically Induced Pounding of El-evated Bridges by Using Magnetorheological Dampers,” Proceedings of the 3rd International Conference on Bridge Maintenance, Safety and Management — Bridge Maintenance, Safety, Management, Life-Cycle Performance and Cost, pp. 685–686, 2006.[7] Ramli, R., Pownall, M., Levesley, M. and Crolla, D.A., “Dynamic Analysis of Semi-active Suspen-sion Systems Using a Co-simulation Approach,” Multi-Body Dynamics: Monitoring and Simulation Techniques-III, pp. 391–399, 2004.[8] Lee, S.H. and Hwang, Y.S., “A Study on a Sce-nario Using the PID Method,” Progress in Nuclear Energy, vol. 51, no. 2, pp. 253–257, 2009.[9] Hong, S. R., Wereley, N. M., Choi, Y. T. and Choi, S.B., “Analytical and Experimental Valida-tion of a Nondimensional Bingham Model for Mixed-mode Magnetorheological Dampers,” Journal of Sound and Vibration, vol. 312, no. 3, pp. 399–417, 2008.[10] Guo, Shuqi, Li, Shaohua and Yang, Shaopu, “Semi-active Vehicle Suspension Systems with Magnetorheological Dampers,” 2006 IEEE Inter-national Conference on Vehicular Electronics and Safety, ICVES, pp. 403–406, 2006.[11] Bouc, R., “Forced Vibration of Mechanical System with Hysteresis,” Proc., 4th Conf. on Non-linear Oscillation, Prague, Czechoslovakia, 1967.[12] Wen, Y. K., “Method for Random Vibration of Hysteretic Systems,” Journal Engineering Me-chanics, ASCE, 102, pp. 249–263, 1976.[13] Spencer, B. F., Jr., Dyke, S. J., Sain, M. K. and Carlson, J. D., “Phenomenological Model of a Magneto-rheological Damper.” Journal of En-gineering Mechanics, ASCE, 123, pp. 230–238, 1996.

Fig. 24: Heeling angle data of handling stability test(red = passive, green = semi-active)

Fig. 23: Handling stability test

Fig. 25: Yaw angular velocity data of handling stability test (red = passive, green = semi-active)

0.00 25.00

2.90

-2.80

Heel

ing

angl

e (d

eg)

Time (s)

0.00 25.00

14.00

-14.00

Yaw

ang

ular

vel

ocity

(deg

/s)

Time (s)

Thomas Dreier, University of Applied Sciences and Arts Northwestern Switzerland, Institute of Automation | CUSTOMER APPLICATION


0 1 2 3 4 5 6 7 8

0.25

0.20

0.15

0.10

0.05

0.00

-0.05

-0.10

-0.15

Stee

r ang

le g

amm

a [w

heel

]

Time [s]

steering head angle 18deg

0 1 2 3 4 5 6 7 8

0.3

0.2

0.1

0.0

-0.1

0.2

-0.3

Stee

r hea

d ga

mm

a [w

heel

]

Time [s]

steering head angle 18deg steering head angle 10deg steering head angle 26deg

0 1 2 3 4 5 6 7 8

Stee

r hea

d ga

mm

a [w

heel

]

Time [s]

Weight comparison normal weightWeight comparison heavy weight

0.25

0.20

0.15

0.10

0.05

0.00

-0.05

-0.10

-0.15

$B_chassis$B_handle

$B_wheel_v $B_wheel_h

$B_wheel_suspension_h$B_wheel_suspension_v

1

1

3

3

3

3

3

2

2

21 4

1 1

1 1

1

1

2

2

2

2

2

2

1 1

2

1


CUSTOMER APPLICATION | Thomas Dreier, University of Applied Sciences and Arts Northwestern Switzerland, Institute of Automation Thomas Dreier, University of Applied Sciences and Arts Northwestern Switzerland, Institute of Automation | CUSTOMER APPLICATION

havior could be examined. At the end of the project, a functionally compatible model of the Peraves AG Monotracer MTE-150 was created, with which the driving behavior of a given driving situation could be simulated. This main goal was broken down into parts: Monotracer stabilization, modeling, and driving dynamics tests of oscillation.

PHYSICAL STABILIZATION THEORY OF A SINGLE-TRACK VEHICLE Vehicle stabilization is an important compo-nent of the dynamic pendulum motion testing on the Monotracer. The basic stabilizing moments of a motorcycle are the result of the centrifugal forces of the spinning wheels. All stabilizing moments can be derived from these forces. The cen-trifugal force results from the inertia and rotational speed of the rotating wheel. In accordance with stabilization theory, a motorcycle moves straight as long as the moments gener-ated through struc-tural measures do not exceed the stabilizing gyroscopic moment.

MONOTRACER MODELThe model for the project was very simple. It was created with the additional

Fig. 2: Steering behavior of the Monotracer on a straight road with released handlebar; test speed 40 km/h

Fig. 3: Monotracer model in SIMPACK

SIMPACK licenses Automotive and Delft MF-Tyre/MF-Swift 6.1.2. The model consists of a chassis similar to a CAD drawing of the original Monotracer, the handlebars which have the form of the original CAD drawing, and the front and rear tires (Fig. 3). The cor-responding Force Elements have also been

included. Two wheel dampers were cre-ated in the model. The wheels also included wheel/ground contacts. For the validation, different lateral impacts and steering angles were simulated. General parameters are defined as part of the modeling process. Among the most important parameters are mass and the inertia of various components. Videos from test drives were analyzed to determine the vehicles movement on a straight road. By the end, a model was created that expresses the driving dynamics characteristics of the examined motorcycle.

DYNAMIC TESTS WITH THE MODELDynamic tests are an important aspect of this project. The questions of where the pendulum oscillations at low speeds come from, or how they can be reduced, can now be studied through various theses and experiments performed on the model. The dynamics of a motorcycle are determined by the overall structure: from the entire system’s geometry, inertia and wheels.

Therefore, the oscillations could be originating from a number of places, and it is unlikely that the properties of each individual piece could

be understood and physically accounted for. For these reasons, a simple model was

“..., measure to suppress the pendulum vibration

could be found.”

Fig. 4: 2D-view of the model

Fig. 5: Influence of the head angle on vibration

Fig. 6: Influence of the total weight on oscillation; the red curve shows the relationship with half of the original mass

created that limited itself on account of the stabilization theory of the steering system and its components.The first experiment studied the influence of the pendulum by changing the steering geometry. With the movements of the head angle, the running of the tires and the entire steering system changed. The studies have shown that by reducing the head angle, pendulum oscillations become smaller, but cannot be eliminated altogether. The origin of the pendulum oscillations that exist at low speeds could not be determined by this investigation. Fig. 5 shows the vibration behavior of various head angles.The second experiment regards the mass distribution of the entire motorcycle. The pendulum oscillations of the Monotracer are, like the centrifugal forces of the individual wheels, dependent on speed. These parallels were studied. The centrifugal force of a wheel is defined at a constant speed, and the motorcycle must be stabilized with this force. If the vehicle is too heavy or the force is too small, the motorcycle will not be entirely stable (Fig. 6). This hypothesis was proved by two studies. For one, the inertia of the wheel was increased; for the other, the total system mass was reduced. Both tests had an eliminating effect on the Monotracer’s pendulum oscillations. RESULTSAll objectives for the project were met. The resulting model can be used for dynamic studies and corresponds to the original Monotracer. As a result of the studies conducted on the model, measures to suppress the pendulum vibration could be found. The increase in wheel inertia is one of the most important measures. Its impact on driving safety must be taken into account, as does the decrease in overall mass, which can be taken into consideration for further developments. In this way, the study demonstrated solutions for improving the dynamic behavior of the Monotracer.

REFERENCES[1] Cocco G.; translated by Schwarz, W.; "Motorrad-Technik pur : Funktion — Konstruk-tion — Fahrwerk", 2001.[2] Bayer B.; "Das Pendeln und Flattern von Krafträdern", Untersuchungen zur Fahrdynamik von Krafträdern unter beso, 1986. [3] SIMPACK AG; "SIMPACK Trainingmanual"; Automotive SIMPACK Training course.[4] Stoffregen J.; "Motorradtechnik. Grundlagen und Konzepte von Motor, Antrieb und Fahrwerk", ATZ/MTZ refenrence book, 2012.

Ready for Print - 03.02.2014 Ready for Print - 03.02.2014

SIMPACK NewsSIMPACK News

INSIDE THIS ISSUE

CONTACTS

INDIAProSIM R&D Pvt Ltd.#4, 1st 'B' Main, 1st 'N' Block Rajajinagar, Bangalore – 560010, IndiaPhone: +91 80 2332 3020 +91 80 4127 7792Mobile: +91 99 7230 4447Fax: +91 80 2332 [email protected], www.pro-sim.com

1. Tire Modeling from Structural Analysis to Real-Time Applications

2. TNO’s MF-Tyre / MF-Swift and the Delft-Tyre Toolchain

3. FTire: High-End Tire Model for Vehicle Simulation in SIMPACK

4. Musculoskeletal Model of Bicycle Pedaling

5. Load Calculations for Wind Turbines

6. Comparison of Simulation and Measurements from On-Track Tests for Model Validation

7. Kinematics Analysis and Design Optimization of Semi-Active Suspension for a Light Bus

8. Modeling, Simulation and Dynamic Analyses of a Closed Single-Track Vehicle

GERMANY (HEADQUARTERS)SIMPACK AG Friedrichshafener Strasse 1 82205 Gilching, GermanyPhone: +49 8105 77266 0 Fax: +49 8105 77266 11 [email protected]

USASIMPACK US Inc.25925 Telegraph Road, Suite 401Southfield, Michigan 48033, USAPhone: +1 248 996 8750Fax: +1 248 996 8930 Mobile: +1 251 923 [email protected]

GREAT BRITAINSIMPACK UK Ltd. The Whittle Estate Cambridge Road, WhetstoneLeicester LE8 6LH, UKPhone: +44 116 27513 13Fax: +44 116 27513 33Mobile: +44 7767 416 [email protected]

FRANCESIMPACK France S.A.S.Immeuble "Le President", 4eme étage40, Avenue Georges Pompidou69003 Lyon, FrancePhone: +33 4 37 91 96 90 Mobile: +33 6 73 88 19 [email protected]

BRAZILVirtualCAE Com. Serv. Sistemas Ltda.Rua Tiradentes, 160 – Sala 22São Caetano do Sul – São PauloCEP: 09541-220, Brasil

Phone: +55 11 4229 1349Mobile: +55 11 9698 [email protected] www.virtualcae.com.br

JAPANSIMPACK Japan K.K.4F Hirakawacho K Bldg.2-4-5 HirakawachoChiyoda-ku Tokyo 102-0093, JapanPhone: +81 3 32 65 78 33Fax: +81 3 32 65 78 [email protected], www.SIMPACK.jp

CHINA

Global Engineering Technology Group (GET Group) Tri-Tower, C Building, Room 1802, 1804Zhongguancun East Road 66 Haidian District, Beijing 100190, P.R. ChinaPhone: +86 10 626 708 [email protected]

TURKEY

RMC MühendislikHavaalani Ic Yolu Caddesi No:3 Teknopark Istanbul 1B47Pendik 34912 Istanbul, TurkeyPhone: +90 216 290 51 72 / [email protected] www.rmc.com.tr

REPUBLIC OF KOREAKostechRm. 804 Nam-Jung City Plaza 1th

760 Janghang-dong ilsandong-guGoyang-si Gyeonggi-do Republic of Korea, zip code: 410-380Jong-Wha, LeePhone: +82 31 903 [email protected], www.kostech.co.kr

SIMPACK News

EDITORIAL, DESIGN & LAYOUT:Steven Mulski, Nicole Blum

©2014 SIMPACK AG. All rights reserved.SIMPACK AG, Friedrichshafener Strasse 1, 82205 Gilching, GermanyPhone: +49 8105 77266 0 Fax: +49 8105 77266 11 [email protected]; www.SIMPACK.com

CIRCULATION: 4.500PUBLICATION YEARS: 1996–2014

All previous SIMPACK News articles can be found at www.SIMPACK.com > downloads > newsletters/articles If you would like to sign up for free delivery of the SIMPACK News please visit: www.SIMPACK.com > downloads > newsletters/articles > subscription

MATLAB and Simulink are registered trademarks of The MathWorks, Inc. Other product or brand names are trademarks or registered trademarks of their respective holders. ©Photography and images are courtesy use of respective authors/companies and may not be copied, printed or otherwise disseminated without express written permission of SIMPACK AG or respective authors/companies.

Ready for Print - 20.02.2014

simpack a friedrichshafener strasse 1 82205 ilching …€¦ · · 2014-03-20simpack a...

Documents