Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Real-time Simulation of a Vehicle Door Locking 2010-01-0666

Published
Mechanism on a Hardware-in-the-Loop Platform 04/12/2010

Suguna Thanagasundram and Ross McMurran

International Automotive Research Centre (IARC), The University of Warwick

R. Peter Jones
School of Engineering, The University of Warwick

Alexandros Mouzakitis
Jaguar Land Rover Engineering Centre

Copyright © 2010 SAE International

availability of the parts, particularly across the potential

ABSTRACT tolerance spread, and associated time and costs. Hence it is
An automotive side door latch release mechanism has been highly desirable to have multi-domain models of
modelled for the locking and unlocking vehicle functionality electromechanical components that can be tested in real-time
in Dymola. The performance of the developed door lock in their system context within an automated test environment
model is evaluated against an existing model of a similar door such as a Hardware-in-the-Loop (HIL) platform. To assess
locking/unlocking system in Stateflow. The model the practicality of such an approach, a side door latch release
performance is also compared with measurements from a real mechanism model has been developed for the locking and
vehicle door latch. The model is converted into a Simulink unlocking vehicle functionality in Dymola.
model and built for a real-time environment such as the
dSPACE target with a fixed step size solver. It is shown that The construction of latching mechanisms vary from one to
a step size as small as 1 ms can be used for real-time another, based on specification such as size, functionality,
simulation without task overrunning in the real-time target. weight and door structure but most door latch modules are
The model is also benchmarked on a multiprocessor setup as kinematically coupled electromechanical systems of rotation
multiprocessor simulators are common in system-level or rotation translation type [1,2,3]. An example of a door
networked Electronic Controller Unit (ECU) testing facilities latch assembly is shown in Figure 1.
for implementing high fidelity closed loop models of
integrated ECUs and actuators. It is proven that the developed
door lock model in Dymola can be built and executed on a
multiprocessor platform and there is much potential on the
use of such models for future work in ECU testing.

INTRODUCTION
In the automotive industry, there is increasing use of electro-
mechanical or “mechatronic” components. To validate such
components the control and response must be assessed taking
into account both its electrical and mechanical behavior
including the interaction with other system components,
environmental performance and tolerance spread. To perform
such tests on physical parts is often not practical due to the
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

NEED OF SIMULATION OF THE

DOOR LATCH
Designing an optimal door latch assembly is a difficult task
as it has to take into account several design considerations.
Firstly the door latch has to be small and compact owing to
the vehicle design and space constraints and has to be limited
to the minimal number of components with the least cost. The
chosen motor needs to have the required power requirement
to drive the connected levers and rods and also overcome
resilience of biasing components like springs and friction
between moving parts. However it is not desirable to put in
Figure 1. Example of a side door latch assembly in a too big a motor as this will only add to weight and cost of the
Ford car (reproduced from [3]). door latch. The inertia of heavy connecting parts need to be
calculated beforehand as this may impede the locking /
Most are either manually operated or power actuated through unlocking operation of the door latch if the motor is not
a motor, consist of a mechanical sub-assembly, a gear train sufficiently powerful and this in turn might lead to erroneous
arrangement, rigid and elastic elements such as articulated operation of the door latch. The material composition,
bars, cams, levers and springs, an actuator in the form of a geometry as well the orientation of the connecting
locking lever, and microswitches through which they provide components has to be designed carefully for reliable
feedback to the door Electronic Controller Unit (ECU). The operation of the door latch.
switches determine the position of the lever at the end of an
actuation and usually have two distinct positions: locked or To the component supplier, the use of a simulation modelling
unlocked. The latch module also works in conjunction with tool such as Dymola [4,5,6] in place of actual prototypes can
the inside and outside door handles via attached rods and enhance the overall design process of the product such as the
cables. automotive door latch. Simulating the door latch instead of
building prototypes can reduce the time and cost required to
This work is concerned with the development of a physical reach the final physical prototype in the development
dynamic model for the locking/unlocking functionality of one lifecycle of the door latch. The inputs used in the Dymola
such electro-mechanical automotive door latching model can be used to predict the response when parameters of
mechanism. The modelled power door actuator comprises of the latch mechanism are changed and these can be used to
two sets of components. The first set of components executes ensure the design of the latch and its electrical control system
Central Door Locking (CDL) functionality and comprises of robustly meet the quality, safety and reliability requirements
a Permanent Magnet Direct Current (PMDC) motor for in the automotive industry. To the OEM (Original Equipment
providing the driving torque and is denoted as the CDL Manufacturer), the primary requirement of a physical model
motor. In addition, the CDL set also comprises of a control is to validate the control strategy robustness in integration
system for regulating the voltage and timing of the opening testing. The door latch is connected to the door ECU which in
and closing actions, a gear reduction stage in the form of a turn has complex dependencies on other ECUs in vehicle
worm gear, a Pinion, a latch release device known as the communication network. If prototypes of the door latches are
LockLever for engaging and disengaging the latch to the not available, the physical model of the door latch can be
vehicle door, a biasing spring and an electrical switch for used in place to validate locking/unlocking functionality.
providing feedback to the door ECU to inform of the current Also a range of parameters can be varied easily in the
state of the latch. The second set of components enables the physical model such as spring stiffness coefficient, motor
superlocking feature in the door lock mechanism. torque etc and the effect this has on the performance of the
Superlocking disables the interior locking control so that the locking/unlocking functionality can be studied.
door cannot be opened from the inside. The second set of
components comprises of another PMDC motor known as the In this paper, the door latch assembly is our primary interest
SuperLock motor which has a lower power requirement due and we study the working principles of the actuator in detail.
to the fewer number of components involved. The other Basically the door latch is the “Locked” state or “Unlocked”
components in this set are a gear in the form of Deadbolt state and it moves between these two states. This movement
Pinion and a Rack for translating the rotational movement of can be created as a result of one of the following actions:
the Deadbolt Pinion into longitudinal movement. torque developed by the power actuator in the form of the
PMDC motors, the movement of the interior door handles or
the movement created by the door keys in the key barrel.
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

The capability of 2 different simulation tools namely: The Dymola model is embedded as a S-function into a
Stateflow and Dymola is evaluated for functional modelling Simulink wrapper model using Dymola's Simulink interface
of the locking/unlocking mechanism in the door latch. block. The Stateflow model of the door latch and the Dymola
Stateflow is an interactive graphical design tool from model of the door latch called from a Simulink block in
Mathworks [7-8] that models and simulates event-driven Matlab are shown in Figure 2. It is the purpose of this work to
systems. Event-driven systems transit from one operating create simulations of the door latch in Dymola and Stateflow
state to another in response to events and conditions. To test which are comparable to real measurements taken from an
the real-time capability, the door latch models in Dymola and actual door latch in a car. Figure 3 shows the measurements
Stateflow are integrated in to the Matlab/Simulink taken from real door latch in a car. The door latch is locked
environment and built for a real-time target such as a by the generation of a 100ms pulse on the CentralLock signal
dSPACE multiprocessor [9] HIL simulator using the Real- by the DDM ECU. The LockFeedback transitions from low
time Workshop of Mathworks [10]. Simulation results are to high (0 to VBat) and the UnlockFeedback transitions from
presented for the door latch model on 4 different platforms. high to low (VBat to 0). VBat is the vehicle battery voltage
They are: and it varies from 12V to 14V depending on the make of the
car and the context of the application it is studied in. There is
1). Door latch model in Dymola (off-line simulation)
a small delay of about 30ms from the time the control signal
2). Door latch model in Stateflow (off-line simulation) is issued to the time the feedback signals change their states.
This time is reflective of the operating time of the CDLmotor
3). Real-time Door latch model in Stateflow
in the door latch. Both the door latch models in Dymola and
4). Real-time Door latch model in Dymola Stateflow are benchmarked to assess their performance in
encapturing this behavior as closely as possible to the real-
The door latch described in this paper has 5 inputs and 2 life door latch and the following section presents both the
outputs. The first three input signals are CentralLock, models in detail.
SuperLock and LockCommon. These are voltage signals
provided to the PMDC motors in the door latch from the <figure 2 here>
networked ECU that controls the door latch known as the
Driver's Door Module (DDM). The other inputs are <figure 3 here>
InteriorLock and InteriorUnlock. These inputs denote the
locking and unlocking action created by the pressing of the DOOR LOCKING MODEL IN
interior door handles of the vehicle. The two outputs from the
door latch are LockFeedback and UnlockFeedback. They
DYMOLA
indicate the “Locked” status and “UnLocked” status The kinematics analysis of each component of the door latch
measured from the microswitches in the door latch system is modelled in Dymola through torque balance and
respectively. The modelled door latch has a superstate known angular equations. Dymola is a flexible tool for modelling
as “SuperLocked”. In this state, when the door latched is and simulating complex integrated multi-physics, multi-
locked, it cannot be unlocked by the interior door handles. engineering systems. In Dymola, components are connected
at a high level using graphical icons that represent models of
The Stateflow model of the door latch has three extra inputs. the components, in such a way as it would be in the real
These are the LockTime and UnLockTime input variables world. The physical couplings between the components are
and denote the time taken to lock and unlock the car achieved by means of connectors and the physical properties
measured from the time the remote key fob is pressed to the of the components are set by initial conditions and parameter
actual time it take to execute these actions. The values of values in dialogue boxes. The reader is referred to an earlier
these inputs are empirically derived from measurements paper which describes this model in detail [11]. The high-
taken from the door latch. The value chosen varies depending level model of the automotive latch mechanism developed in
on the dynamics of the system and is determined from Dymola is shown in Figure 4. The components shown in the
parameters like power developed in the motor, the stiffness in Dymola model are the CDL motor, Wormgear, Pinion,
springs used and gear ratios used in the door latch assembly. LockLever and springs S1 and S2 for CDL locking and two
The last extra input of the Stateflow model is the KBunlock subsystems for superlocking and interior locking/unlocking.
which denotes the action of unlocking the car manually by The CDL motor powers the movement of the latch from the
turning the car's key in the key barrel. The Dymola door latch locked to unlocked position. A lossy worm gear (consisting
model is not modelled with the LockTime and UnLockTime of the Worm and the WormWheel) is used in transmission of
input variables. The reason for not using these variables for motion from the CDL motor to the articulated LockLever and
the Dymola model will be explained in a later section. The pinion. The worm gear converts the CDL motor's high speed
KBunlock input variable is also not used in the Dymola door and low torque to match the requirements of low speed and
latch model as the feature is still not implemented at the high torque of the articulated levers. The gear ratio of the
moment. worm gear determines the relationship between the input and
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Figure 2. (a) Simulink Block calling the Door Latch model in Stateflow. The block has 8 inputs and 2 outputs (b) Simulink
Block calling the Door Latch model in Dymola. The block has 5 inputs and 2 outputs.

Figure 3. Measurements taken from a real door latch in a car.

output rotational speed and the relationship between the input clockwise to anti-clockwise, the referenced fixed point (from
and output torque. There are two torsional springs in the which the angle of deflection is measured) changes abruptly.
modelled door latch mechanism. A preloaded torsional spring This behavior is modelled as a discontinuity in Dymola.
S1 with damping and spring stiffness exists between a Torsional spring S1 is in contact with the WormWheel which
variable Fixed Point and WormWheel. A normal preloaded in turn is connected to the Pinion that is in turn connected to
torsional spring S2 with spring stiffness is connected between the Lever. The WormWheel, Pinion and LockLever behave
the Fixed Point and the Lever. like a simple gear train in the modelled latch. Initial input
values to the Dymola components are estimated by simple
Torque is transferred to torsional spring S1 from the CDL calculations. The original model has 1248 unknowns. After
motor via the worm gear. Torsional spring S1 is in direct Dymola's elimination of constant and alias variables at
contact with the WormWheel and when the WormWheel translation, 705 non-trivial variables remain. The model has
rotates, the spring is deflected in the positive direction, 14 continuous time states and 211 time- varying variables.
increasing the angle of deflection from the centre of rotation
and decreasing its diameter. The torsional spring S1 has two <figure 4 here>
fixed points depending on the direction of rotation of the
worm gear. When the WormWheel changes its direction from
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Figure 4. Top-level door latch model in Dymola

The door lock model in Dymola includes non-linear effects very small time step, as small as 0.1 ms, has to be used to get
such as elasticity of the parts, sliding friction contact and accurate results.
backlash between the gear teeth, causality changes (e.g. hard
stop) and has many state events. It has both very fast and very The dynamics of the off-line door latch model in Dymola are
slow dynamics and switching points in time (discontinuities) verified, the behavior during a lock/unlock/superlock
depending on the direction of motion of the worm gear. operation is studied and results are depicted in Figure 5. The
Hence the lock model in Dymola is a hybrid model results are obtained using the variable DASSL solver. For
containing continuous systems and discrete systems as well off-line simulation of the Dymola model with the DASSL
as a model with stiff components. Tuning the model for fixed solver, 0.265s of CPU time is used on an Intel Duo Core
step integration is a major task. Instead it is computed by processor with a speed of 2.20 GHz and 2GB of random
using the default variable step integrator such the Differential access memory. During the simulation, 200 state events
Algebraic System Solver (DASSL) solver in Dymola with a occur. At 0.1s, a VBat voltage of 12.5V (with 100ms
numerical accuracy of tolerance set to 0.001 for the offline duration) is fed to the CDLMotor via the CentralLock signal.
experiment. For the off-line model, a variable solver such as The LockFeedback signal changes from “0” to “1” at around
DASSL is used as other solvers, like Runge-Kutta methods, 0.15s indicating the locked state. At 0.6s, a -VBat (with
or fixed step solvers are less suitable for such kind of a 100ms duration) signal is applied to the CDL motor via the
simulation task and can lead to more inexact results LockCommon signal for the unlock operation. The
especially if the model is stiff (probably due to a lot of LockFeedback signal changes from “1” to “0” at around
events). It was possible to simulate the off-line model of the 0.67s indicating the unlocked state. At 1.1s, VBat is again fed
door latch in Dymola with a fixed step solver like Euler but a to the CDLMotor for 100ms via the CentralLock signal and
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

the LockFeedback changes its state correspondingly. At 1.5s, the sill button in interior handle of the vehicle, what happens
the door latch is superlocked by applying the VBat voltage next or at the press of the lock/unlock button in the key fob
(with 100ms duration) to the SuperLock motor via the and what are the resulting actions. The time-based behavior
SuperLock signal. This causes the deadbolt Pinion to rotate depends on the timing of actions. i.e. on ‘when’ the lock
and this translates to a longitudinal movement by the Rack. button in the key fob is pressed, it is required to allow an
The rack moves from the 1st position to the 2nd position and actual further 50 ms or 100ms to execute the locking action.
physically blocks the LockLever from rotating anti- The statechart has 2 main superstates; Unlocked and Locked.
clockwise. At 1.8s, it is attempted to manually unlock the Within the Locked state, the door latch can be in a
door latch through the interior release handles by applying a Singlelocked or Superlocked state. A few minor states such
short pulse output of VBat through the as UnlockReady, LockReady, UnSuperLockReady, and
interiorLockandUnlock subblock. However the door latch SuperLockReady are also encapsulated to allow enough time
remains in the locked state as it has been superlocked. The to elapse before the door latch transits from one state to
simulation is stopped at 2.0s. another. Hence two other extra inputs variables LockTime
and UnLockTime are used in the Stateflow model for the
door latch. These variables were not required in the Dymola
model as Dymola itself is an acausal tool and the time taken
to execute a behavior is determined from only the complexity
of the model, the integration algorithm (solver) used and the
simulation time step. The time taken to execute the “lock”
and “unlock” functionality is determined from the Dymola
model itself, but this is not the case for the Stateflow model.
Such parameters had to be explicitly stated as dynamics of
the model depend on event based behavior which in turn also
depends on the discrete time lapsed. The Stateflow diagram is
shown in Figure 6. The offline Stateflow model was run in
Simulink with a fixed step solver Euler with a time step of
1ms for 5s (Figure 7). The states in the model transition from
one state to another depending on the time and condition of
the input variables.

For the Stateflow model of the door latch, initially it is in the

unlocked state. This is denoted by the LockFeedback being
“0” and the UnlockFeedback being “1”. The LockFeedback
and UnLockFeedback outputs from the door latch model are
the inverse of each other. The LockTime and UnLockTime
are taken to have a duration of 50ms. At 0.5s, a 100ms VBat
voltage is issued via the DDM to the CDLmotor via the
CentralLock signal. The door latch transits from the unlocked
to the locked state and this can be seen from LockFeedback
Figure 5. Off-line door latch model in Dymola changing from “0” to “1” and UnlockFeedback changing
demonstrating the locking/unlocking/superlocking from the “1” to “0”. At 1s, the interior_unlock handle is
functionality. Simulation results are obtained using the pulled (shown as a very pulse with a very short duration) and
DASSL solver for a time duration of 2s. the door latch changes from the locked to the unlocked state.
At 1.5s, the door latch is locked from the interior door
handles and the status of the door latch changes
STATEFLOW DOOR LOCKING correspondingly. At 2s, the latch is superlocked. This is
achieved by issuing a 100ms VBat pulse on the SuperLock
MODEL signal. At 2.5s, the door latch is unlocked from the interior
A Stateflow model has been developed to describe the door handle but remains in the locked state. At 3s, the door is
“event-based” and “time-based” behavior of the door latch. unlocked by the CDL motor by issuing a 100ms signal on the
The described Stateflow model is an improved version of the LockCommon signal. The door latch is again locked from the
Stateflow model presented in the piece of work by Kendall interior door handle at 3.5s. At 4s, the door latch is unlocked
and Jones [12]. As described in this paper, the “event-based” by pressing the interior door handle. The door latch unlocks
behavior of the door latch can be modelled in states. The this time it is not superlocked. At 4.25s, the door latch is
“event-based” behavior depends on causal relationships locked by the CDL motor. At 4.75s, the door latch is
between external events and actions such as at the press of
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Figure 6. Stateflow Diagram showing the “state” transition behavior of the door latch.

unlocked by turning the key in the key barrel. The simulation use a powerful computational platform such as the HIL
is stopped at 5s. platform to provide the sufficient computing power for real-
time simulation and to satisfy deterministic real-time
<figure 6 here> constraints. The use of a HIL simulation platform to test and
analyze automotive electronic control systems has been well
<figure 7 here> documented by many researchers [13,14,15,16,17,18]. A HIL
platform is a simulation testbed that is used in the
EVALUATION OF REALTIME development and testing of real-time target embedded
controllers in the ECUs and actuators such as the door latch.
BEHAVIOUR OF DYMOLA DOOR The main function of a HIL system is to emulate the
LATCH MODEL dynamics of the real process, including the behavior of the
sensors and actuators, in order for the physical ECU to think
For system verification and functional testing purposes, it is
that it is part of the real vehicle.
necessary to integrate the developed plant models of the door
latch with the DDM ECU controller. It becomes necessary to
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Figure 7. Off-line door latch model in Stateflow demonstrating the locking/unlocking/superlocking functionality. Simulation
results obtained using the Euler solver for a time duration of 5s.

The benefits of using a HIL platform for automated testing as are connected to and from the I/O ports of the DDM ECU
opposed to the traditional way of performing testing with the hardware interface through the Real-Time Interface (RTI)
actual hardware are reduced development costs and blocksets specified by dSPACE. The models and application
timescales, availability of the system for 24 hours and safer software are compiled, built and loaded into both master and
testing conditions because of the ability to simulate operating slave servers in real-time by using the Real-time Workshop
conditions like fault and damaged conditions. Functionality from Mathworks and the dSPACE target compilers.
testing can begin earlier in the design process with models of
the ECUs and actuators even before the actual hardware The door latch models in both Dymola and Stateflow are
become available. Faults found earlier in the design phase are simulated in the real-time environment and are shown in
cheaper to fix as the extra software iterations from the Figure 8. The inputs to the models CentralLock,
supplier are expensive once the hardware have been LockCommon and SuperLock are captured through Analog-
prototyped. Testing procedures can be automated making the to Digital (ADC) card in the real-time target from the real
testing process efficient and cost effective. door latch. The outputs from the model LockFeedBack and
UnLockFeedBack are fed back in closed-loop control to the
In this work, a HIL platform having two full-size HIL DDM ECU via digital inputs in the application studied. The
simulators (servers each hosting dSPACE DS1006 real-time outputs from the real door latch are also captured as digital
processor board running the simulation models and the inputs. The choice of which set of inputs to be fed back to the
various dSPACE real-time I/O cards) is used. The dSPACE DDM ECU is determined by a multiport switch. If the value
DS1006 processor is based on a powerful 64-bit 2.6 GHz of the multiport switch is set to 1, the simulated outputs from
microprocessor with 1MB of Level 2 cache memory. Each of the Dymola door latch model are fed back as inputs to the
the microprocessors run on a real-time operating system and DDM ECU. If this constant block is set to 3, the simulated
has a scheduler running multi-timer tasks with triggering outputs from the Stateflow door latch model are fed back as
capabilities. The main processors on each rack are inputs to the DDM ECU. If it is desired to send out the real
interconnected via a high-speed Gigalink (1.25 Gbit/s) feedback from the real door latch, the value is set to 2. The
connection. The variables of the simulated door latch model DDM ECU is also physically hardwired to the real door latch
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Figure 8. An overall diagram showing the simulated door latch models in the real-time HIL platform and connections from the
real DDM ECU to real door latch.

as it would be in a real car. The motor drive circuit in the Hence a suitable simulation time step had to be determined
DDM ECU provides the necessary voltage and current to the for real-time simulation and analysis of the Dymola door
PMDC motors in the real door latch to perform the locking/ latch. It is desirable to minimize the simulation time step to
unlocking/superlocking actions. Though both the door latch achieve highest accuracy as possible within the computing
models in Dymola and Stateflow were simulated in real-time, capacity of the HIL processor. It was decided that the real-
for purposes of this study, only the simulation process of the time code of the simulated Dymola and Stateflow latch model
Dymola model is explained in greater detail in this paper will be run with a timer task with a step size of 1 ms as this
here. was the time step used for other plant models in the vehicle
testing.
<figure 8 here>
For the Dymola model, apart from being a fixed time step
For compilation of the Dymola door latch model for real- solver, the solver also has to have inline integration to be
time, a suitable solver had to be chosen. A variable step used for numerical integration in the HIL platform [19-20].
solver like DASSL used in the off-line simulation of the Inline integration approach is new concept in Dymola and is
Dymola model was not used as there is a possibility that the exploited by the tool to reduce the size of the-non-linear
length of time taken for computer execution for the problem. This technique merges the discretization formulas
calculations in a given integration frame will be bigger than of the integration algorithm along with the equations of
the used time step. This in turn means that the simulation simulation model as a separate or an inlined algorithm prior
output at the end of that time step will be inaccurate and will to the symbolic preprocessing of the resulting Differential-
cause task overrunning in the HIL platform. Also HIL Algebraic Equation (DAE) system. This enables real-time
platforms only work with a fixed step size. This is a simulation of potentially stiff parts in the Dymola model so
mandatory requirement in HIL platforms as it is required to that a fixed-step solver like Euler can be used and better
maintain compatibility with fixed sample rates when dealing accuracy can be obtained with a larger step size.
with real-time inputs and outputs.
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Table 1. Real-time Dymola model of door latch is run with different solvers

Simulation of stiff model components with slow dynamics in And in the real-time target, the code for the master and slave
Dymola is difficult and sometime it is advantageous to use processor had to be compiled and built separately.
implicit methods of integration which allow bigger timesteps
to be used but lead to a set of non-linear equations [21]. It CONCLUSIONS
was also investigated whether to use explicit or implicit
integration or to use both (which is referred as mixed-mode In this paper, we have looked at two different simulation
integration) in Dymola. Results for the Dymola door latch tools namely Dymola and Stateflow for modelling a multi-
model with the various solvers and with a time step of 1ms domain door latch mechanism. Each of the tools has its
are listed in Table 1. The turnaround time is the time taken strength and weaknesses. Dymola software is useful for
for the dSPACE target to execute the simulation model. It seamlessly modelling and simulating the multi-body multi-
was found that only explicit integration methods were domain latch mechanism. In Dymola, the modelling and
suitable and when implicit methods were chosen, they simulation are closer to the system that a user would use in
typically caused task overrun in the timer of the HIL platform real life and the behavior is modelled as equations. Dymola
and made the simulation stop abruptly (the turnaround time also adopts non-causal modelling approach.
was more than the stepsize specified for the model). In our
case the explicit inlined Euler method and explicit inlined The model in Stateflow is event driven, based on timing
Runga Kutta solvers gave accurate results. detailed in specification and condition of the input variables.
This is the reason why the specific model variable such as
<table 1 here> Locktime and Unlocktime had to be included in the Stateflow
model of the door latch. There was no physics modelled in
A Graphical User Interface (GUI) was made with dSPACE the Stateflow model and effects such as tolerance,
ControlDesk for visualization and for monitoring simulation temperature, friction voltage were not included. In Stateflow,
parameters of the online real-time door latch models in causality has to be determined and set of input- output
Dymola and Stateflow. Results are shown in Figure 9 for one relationships clearly defined.
instance when the locking functionality was implemented. At
0.1s, a 100ms VBat pulse is issued via the CentralLock The semantics used in Dymola were different to Stateflow.
signal. Both the real-time models of the door latch in Dymola The Dymola model is closer to a real-life door latch as
and Stateflow transit from the unlocked to the locked state parameters such as torque, angle and resistance in the
and this can be seen from LockFeedback changing from “0” components could be measured and benchmarked. In terms of
to “1” and UnlockFeedback changing from the “1” to “0”. modelling effort, the Dymola model took a greater time and
For the Stateflow model, the change is immediate due to the was more difficult to construct compared to the Stateflow
event-based behavior of the Stateflow model but for the model. Different skills were needed compared to those
Dymola model the change is delayed by 30ms which is more existent in control engineers and a different mindset and way
realistic to the behavior exhibited by the real door latch. The of thinking was required to construct the semantics in
unlocking functionality is very similar to the locking Dymola as it was based on acausal architecture, but once
functionality. physics of objects were understood, it was a straightforward
process to implement the model in Dymola. On the other
<figure 9 here> hand, the simplicity of the Stateflow tool made it easier to
construct the door latch model in a short period of time.
For the real-time Dymola model, it was found that the
Dymola model could be built for a multiprocessor platform. The real-time behavior of these models was also investigated.
Also Simulink blocks with multiple Dymola blocks could be Building the Stateflow model of the door for a real-time
built and run. The only drawback was that Dymola block platform was a straightforward and standard process as it was
with Animation and multi body systems could not be built. integrated with the Mathworks environment. However in
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

Figure 9. The real-time Dymola and Stateflow models running on the dSPACE multiprocessor platform and exhibiting the
locking unlocking functionality.

Dymola, building the models for a real-time platform was a Dymola door latch model is more appropriate. For example,
more complex process and an investigation with several in the case of engineers trying to develop control algorithms
solvers had to be carried out before choosing the right solver in the absence of the real part, the Dymola door latch model
and right methods of integration. It was also necessary to use would be more appropriate as it provides better fidelity,
inline integration. Moreover Dymola lacks its own solver for behavioral information and dynamic characteristics of the
the real-time implementation and relies on the solvers of physical behavior as would the real door latch. Whereas the
Simulink. However once the models were built and compiled Stateflow model of the door latch will be adequate when
for a real-time platform in Dymola, repeating the same validating ECU behavior or when doing ECU network
process for further modifications in the model was a standard integration testing, as the objective here is to find out the
procedure. In our study, we managed to build the model correctness of the overall ECU software behavior.
successfully on a multiprocessor real-time platform and run it
without any major issue. REFERENCES
It can be argued that in deciding whether to use Dymola or 1. Bur, A.W., Dierauer, P.P., and Ricks, L.F., “Honeywell's
Stateflow for modelling, with either simulation tools, the Automotive Door Latch Design is Ideal for Corporate Latch
model can only be as good as the upfront analysis and design Strategy,” SAE Technical Paper 2003-01-1190, 2003.
work done for the desired system behavior. It is believed that 2. Gilberg, A., Marcosky, J., Sherman, L., and Clarke, R.,
the timing discrepancies seen in Stateflow model can be “Door Latch Strength in a Car Body Environment,” SAE
rectified if a higher fidelity model is used and if description Technical Paper 980028, 1998.
of the system behavior is correctly interpreted and translated
into Stateflow notation. Hence there is no such thing as which 3. Udriste, D.I. and Negrus, E.M., “Construction and
is better tool, but depends on the need of the application and Kinematics of Automotive Side Door Latch Mechanisms,”
the accuracy of the system model which needs to be captured. SAE Technical Paper 2005-01-0881, 2005.
In the application studied, the choice of the door latch model 4. Dynasim AB, Dynamic Modelling Laboratory, http://
to use relates to how they will be used. For simple basic www.dymola.com/, accessed: 30 September 2009.
testing with a lower effort, the Stateflow model of the door
latch may be adequate but if a more in-depth performance 5. Modelica Association. Specification, Tutorials, http://
testing on robustness studies is to be done, the higher fidelity www.modelica.org/, accessed: 30 September 2009.
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

6. Tiller M., Introduction to physical modelling with Automotive Systems”, Proceedings of the 10th Mechatronics
Modelica, Kluwer, 2001. Forum Biennial International Conference, MX2006,
Philadelphia, USA, 2006.
7. The MathWorks, Inc., Natick, MA, 2004. http://
www.mathworks.com accessed: 30 September 2009. 19. Elmqvist H., Mattsson S.-E., Olsson H., “New Methods
8. The MathWorks, Inc., Stateflow Technical for Hardware-in-the-Loop Simulation of Stiff Models,”
Documentation, http://www.mathworks.co.uk/products/ Proceedings of the 2nd Intemational Modelica Conference,
stateflow/technicalliterature.html accessed: 30 September pp. 59-64, 2002.
2009. 20. Elmqvist H., Cellier F., Otter M., “Inline Integration, A
9. dSPACE digital signal processing and control engineering new mixed symbolic/numeric approach for solving
GmbH, dSPACE GmbH, http://www.dspaceinc.com, differential-algebraic equation systems”, Proceedings of
accessed: 30 September 2009. European Simulation Multiconference, Prague, pp XXIII-
XXXIV, June 1995.
10. Hanselmann, H., “Automotive control: From concept of
Experiment to Product”, Proceedings of the IEEE 21. Soejima S., “Examples of usage and the spread of
International Symposium on Computer-Aided Control Dymola within Toyota”, Modelica Workshop 2000
System Design, pp 129-134, 1996. proceedings, pp 55-59, 2000.
11. Thanagasundram Suguna, McMurran Ross, Dhadyalla
Gunwant, Jones R. Peter and Mouzakitis Alexandros, CONTACT INFORMATION
“Dynamic Analysis of an Automotive Door Latching Dr. Suguna Thanagasundram is working as a Research
Mechanism through the use of Dymola/Modelica”, 7th Fellow in the “Evolutionary Validation of Complex Systems”
International Conference on Manufacturing Research project at the International Automotive Research Centre
(ICMR09), University of Warwick, UK, September 8-10, (IARC), based in the University of Warwick. She has a PhD
2009. in Engineering (in the area of fault prediction and diagnostics
12. Kendall, I.R. and Jones, R.P., “Investigation into the use using Autoregressive (AR) modelling techniques) and a MSc
of hardware-in-the-loop simulation testing for automotive with distinction in Information & Communications
electronic control systems”, Control Engineering Practice, v Engineering from University of Leicester, UK. Her
7, n 11, pp 1343-1356, 1999. undergraduate degree in electrical engineering is from the
National University, Singapore. At IARC, she has worked
13. Li, J.; Yu, F.; Zhang, J.W.; Feng, J.Z. and Zhao H.P., with a number of leading automotive OEMs in a variety of
“The rapid development of a vehicle electronic control hardware-in-loop simulation projects involving HIL
system and its application to an antilock braking system platforms such as dSPACE/add2 and system modelling tools
based on hardware-in-the-loop simulation”, Proceedings of like Matlab, Simulink, Stateflow and Dymola.
the Institution of Mechanical Engineers, Part D: Journal of
Automobile Engineering, v 216, n 2, p 95-105, 2002. Her interests are primarily in the areas of digital signal
14. Gietelink, O., Ploeg, J., De Schutter, B. and Verhaegen, processing, system identification, application of real-time
M., “Development of advanced driver assistance systems platforms using hardware-in-loop and rapid prototyping
with vehicle hardware-in-the-loop simulations”, Vehicle systems, model-based design and advanced physical
System Dynamics, v 44, n 7, pp 569-590, 2006. modelling.
15. Lee, K.C., Jeon, J.W.; Hwang, D.H., Lee, S.H. and Kim, s.thanagasundram@warwick.ac.uk
Y.J., “Development of antilock braking controller using
hardware in-the-loop simulation and field test”, Proceedings
of the IECON (Industrial Electronics Conference) 2004, v 3, ACKNOWLEDGMENTS
pp 2137-2141, 2004.
This work was conducted as part of the Evolutionary
16. Tartt, C.J. and Moskwa, J.J., “A hardware-in-the-loop Validation of Complex Systems (EVoCS) project and
transient diesel engine test system for control and diagnostic supported by the UK Technology Strategy Board (TSB)
development”, Proceedings of the ASME Dynamic Systems under the Project No: TP/4/VCS/6/I/20027.
and Control Division, v 70, pp 255-261, 2002.
17. Isermann, R.; Schaffnit, J.; Sinsel, S., “Hardware-in-the-
loop simulation for the design and testing of engine-control
systems”, Control Engineering Practice, v 7, n 5, pp 643-653,
1999.
18. Mouzakitis, A., Humphrey, R., Bennett, P. and Burnham,
K.J., “Development, Testing and Validation of Complex
 Downloaded from SAE International by University of British Columbia, Monday, July 30, 2018

The Engineering Meetings Board has approved this paper for publication. It has Positions and opinions advanced in this paper are those of the author(s) and not
successfully completed SAE's peer review process under the supervision of the session necessarily those of SAE. The author is solely responsible for the content of the paper.
organizer. This process requires a minimum of three (3) reviews by industry experts. SAE Customer Service:
Tel: 877-606-7323 (inside USA and Canada)
All rights reserved. No part of this publication may be reproduced, stored in a
Tel: 724-776-4970 (outside USA)
retrieval system, or transmitted, in any form or by any means, electronic, mechanical, Fax: 724-776-0790
photocopying, recording, or otherwise, without the prior written permission of SAE. Email: CustomerService@sae.org
ISSN 0148-7191 SAE Web Address: http://www.sae.org
Printed in USA
doi:10.4271/2010-01-0666