Proceedings of the ASME 2009 Dynamic Systems and Control Conference DSCC2009 October 12-14, 2009, Hollywood, California, USA
DSCC2009-2709
DEVELOPMENT OF AN INTERNET-DISTRIBUTED HARDWARE-IN-THE-LOOP SIMULATION PLATFORM FOR AN AUTOMOTIVE APPLICATION Tulga Ersal 1 Jeffrey L. Stein 1
Mark Brudnak 2 Zoran Filipi 1
Ashwin Salvi 1 Hosam K. Fathy 1,*
1
2
Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI, USA The US Army Tank-Automotive Research, Development and Engineering Center, Warren, MI, USA * Corresponding author:
[email protected]
ABSTRACT This paper summarizes efforts to integrate, for the first time, two geographically-dispersed hardware-in-the-loop simulation setups over the Internet in an observer-free way. The two setups are the engine-in-the-loop simulation setup at the University of Michigan (UM) in Ann Arbor, MI, USA, and the driver-in-theloop ride motion simulator at the US Army Tank-Automotive Research, Development and Engineering Center (TARDEC) in Warren, MI, USA. The goal of this integration is to increase the fidelity of experiments and to enable concurrent engineering. First, a model-based simulation of the setup is utilized to analyze the effects of variable delay, an intrinsic characteristic of the Internet, on the integrated system, particularly in terms of stability, robustness, and transparency. Then, experiments with the actual hardware are presented. The conclusion is that the two pieces of hardware can indeed be integrated over the Internet without relying on observers in a stable and subjectively transparent manner, even if the nominal delay is increased by four times.
if the setups are geographically dispersed, bringing them together and establishing a physical connection may be infeasible. In that case a virtual coupling can be created through a communication medium. Different communication mediums are available, and due to its prevalence the Internet is an attractive choice. Using the Internet as the communication medium, however, also creates some challenges due to the Internet’s inherent delay, jitter, and loss. Much of the delay is related to routing and processing in the network. Jitter means that the delay is variable, and loss occurs because not every packet sent through Internet necessarily arrives at its destination. Although the latter can be mitigated by transport-level protocols like TCP, such protocols increase the delay and are therefore not considered in this work. These characteristics of the Internet jeopardize system stability, robustness, and transparency – a measure of closeness to ideal coupling. Thus, maintaining stability, robustness, and transparency despite delay, jitter, and loss is a major challenge in using the Internet to couple HILS setups.
Keywords: real-time hardware-in-the-loop simulation, Internet-distributed simulation, stability, transparency
Within the broader context of bilateral communication over a delayed channel, the literature presents many methods to handle the communication and discusses in detail their advantages and disadvantages. For presentation purposes, these methods can be classified into four categories: direct transmission, prediction/filtering, passivity, and transparency optimization.
I. INTRODUCTION Hardware-in-the-loop simulation (HILS) provides a bridge between physical prototyping and virtual experiments by uniquely combining their advantages and allowing for cost-effective highfidelity experiments [1]. It strongly promotes concurrent system engineering and has therefore become indispensable in many application areas, such as automotive [2], aerospace [3], manufacturing [4], robotics [5], and defense [6].
The simplest communication design is the direct transmission of coupling variables. In this type of architecture the coupling variables are exchanged directly over the communication channel without any transformation, filtering, or control. The advantage of this architecture is its simplicity; whereas its disadvantage is that it does not guarantee stability or transparency, so increasing time delays in the communication medium may lead to a poor
A natural evolution of the basic HILS idea is to integrate multiple HILS setups to fully exploit the benefits of HILS [7]. But
1
Copyright © 2009 by ASME
Griffiths et al. proposed the distortion metric to quantify transparency in an haptic device [24]. They framed the haptic rendering problem as a general control configuration, in which the performance variable is different than the feedback variable, and formulated the trade-off problem between stability robustness and transparency using the fundamental limitations theory for the general control configuration [24]. Transparency was explicitly taken into account for control design by other researchers as well, e.g. in the model-based control scheme proposed by Yokokohji and Yoshikawa [25], or in the telefunctioning framework of Kazerooni and Moore [26].
performance and even to instabilities [8]. One variation of the direct transmission approach is the event-based framework proposed by Xi and Tarn [9] and further developed by Elhajj et al. [10]. The premise of this framework is to find a variable s, called the event reference, that is a monotonically increasing function of time and is the only independent variable for all the signals in the system. Then, under certain conditions, the event-based framework may have guaranteed stability, event-transparency, and event-synchronization, independent of the time delay. Prediction/filtering methods rely on observers to overcome the challenges posed by time delay. Within the specific context of Internet-distributed HILS, previous work by the US Army TankAutomotive Research, Development and Engineering Center (TARDEC) proposed an observer-based method to achieve a stable integration. Specifically, TARDEC researchers successfully integrated a ride motion simulator in Warren, MI, USA, with a hybrid power system simulator in Santa Clara, CA, USA [11-13]. This integration relied on observers on both sides, i.e., each side had a model of the other side and interacted with the other side through these models [11]. The integration ensured, through feedback control, that the states of the observers remained close to the actual states. Different feedback control strategies were employed and analyzed, including PI [11], sliding mode, and H∞ control [12]. Duty cycle experiments have been performed successfully with this setup [13].
These four classes of methods have been extensively studied within the context of telerobotics, but their application to Internetdistributed HILS is limited to the observer-based approach proposed in [11-13]. In this paper we report the first efforts on achieving a stable Internet-distributed HILS using direct transmission, i.e., without relying on observers. Specifically, we adopt an event-based communication, primarily because of its ease of practical implementation, as will be discussed later in detail. The ultimate goal of this work is to investigate the fundamental limits of an observer-free solution to Internetdistributed HILS and how to push those limits, but this paper focuses only on investigating the viability of an observer-free solution. The two HILS setups of interest in this work are the enginein-the-loop simulation setup at the University of Michigan (UM) in Ann Arbor, MI, USA, and the abovementioned driver-in-theloop ride motion simulator at the TARDEC Simulation Laboratory in Warren, MI, USA. The UM setup uses a 6L V8 diesel engine in combination with a vehicle dynamics and driveline model [27]. The TARDEC setup, on the other hand, has a human operator driving a virtual vehicle through a ride motion simulator. It is desired to couple UM’s engine with TARDEC’s simulator to enable the operator to drive a virtual car with a real engine, thereby increasing the fidelity of experiments and enabling concurrent engineering.
The passivity-based framework, on the other hand, offers an attractive guarantee of stability and is a widely used method in telerobotics. There are different ways of achieving passivity for communication mediums with constant time delay, including adding damping to direct transmission [8], using wave or scattering variables as communication variables [8, 14], or using PD control [15]. However, when the delay is variable, as is with the Internet, more sophisticated methods are necessary to guarantee passivity of the communication channel. Some methods include employing energy-conserving filters [16], time-variable gains if a bound on the rate of change of the delay is known [17], or Smith predictors in conjunction with filters to predict and regulate waves [18]. These methods have been further extended to include estimating [19] or predicting delays [20]. Another approach is to keep the apparent time delay constant through the virtual delay method and utilizing the methods for constant delay [21].
With this motivation, Section II of this paper first presents a model-based study of the adopted observer-free integration approach. The models are described in detail, the adopted eventbased communication is described, and simulation results are given along with interpretations for stability, robustness, and transparency. Section III presents the experimental part of this work. The two pieces of hardware that are integrated are described, the experiments are explained, and the results from the Internet-distributed HIL experiments are given and interpreted. The conclusions are given in Section IV.
Even though passivity-based communication methods have the attractive guarantee of stability, this guarantee may come at the expense of transparency. Thus, achieving stability robustness and transparency simultaneously presents a trade-off problem as first recognized by Hannaford [22] and formalized by Lawrence [23]. Lawrence also proposed a metric in a teleoperation device and showed that transparency can be improved significantly by relaxing the goal of having infinite stability robustness, using all four information channels between the master and slave, and optimizing the communication architecture for transparency [23].
II. MODEL-BASED EVALUATION OF STABILITY, ROBUSTNESS AND TRANSPARENCY Before the human drivers and hardware are put at risk, it is desired to perform simulation studies that are entirely based on models to see if a stable integration of the two HILS setups over
2
Copyright © 2009 by ASME
In conjunction with the map, the engine model includes models for the engine inertia, idle controller, and fuel controller. The engine inertia is a small inertia element that takes the load torque and the engine torque as inputs and determines the engine speed. A gain is used to scale the engine torque and simulate engines with more or less power.
the Internet could be achieved using the adopted event-based communication. This section first describes the models considered in this part of the study, describes the communication architecture, and then presents the simulation results. System model Figure 1 illustrates a high-level overview of the HILS setup model considered in this part of the study. The major subsystems are denoted as Driver, Vehicle, Internet, Drivetrain, and Engine, and the signals exchanged between these subsystems are also shown in the figure. These subsystems will be described in detail next.
The idle controller is a PI controller with saturation and antiwindup, and is activated when the throttle demand from the driver falls below 11% to maintain an engine idle speed of 750 RPM. The fuel controller fulfills two purposes. First, it implements the turbo lag as a first order system of the form k / (τ s + 1) . The input to the turbo lag is the difference between the maximum fuel rate for the given engine RPM and the naturally aspirated fuel rate, both determined experimentally. The output of the turbo lag is added to the naturally aspirated fuel rate. This sum is then compared to the maximum possible fuel rate for the given engine RPM and throttle, and the minimum of the two is taken as the unadjusted fuel rate. Second, the fuel controller monitors the unadjusted fuel rate and adjusts it if the engine speed falls below 650 RPM or exceeds the maximum rated speed of 3300 RPM to bring the speed back to the desired operating region.
drive cycle
Driver brake
throttle
vehicle speed
UM
shaft torque
shaft torque
Vehicle
Internet shaft speed
pump speed
Drivetrain shaft speed
TARDEC
Engine load torque
throttle
The drivetrain model includes the torque converter, transmission, and shift logic. The torque converter model is a static model that takes pump and turbine speeds as inputs and generates pump and turbine torques according to the equations
FIGURE 1. OVERVIEW OF THE UM-TARDEC HILS SETUP MODEL
Server side: UM. The server side employs a map-based engine model as shown in Fig. 2. The map is a static map obtained experimentally from the physical 6L V8 diesel engine employed in the UM setup. The input to the map is the fuel rate and the engine speed, and the output is the engine torque. Fig. 3 shows the engine map.
2
τ pump τ turbine
scaled torque
Inertia load torque
torque
K
Engine Map
fuel rate
Fuel Controller
Torque gain
throttle command
(1)
where ωr = ωturbine / ω pump is the speed ratio between turbine and pump speeds, κ (ωr ) is a piecewise function approximating a desired capacity factor curve, and α (ωr ) is a piecewise linear function approximating a desired torque ratio curve.
engine speed pump speed
⎛ ω pump ⎞ = ⎜⎜ ⎟⎟ sign (1 − ωr ) ⎝ κ (ωr ) ⎠ = α (ωr )τ pump
Idle Controller
throttle demand from driver
The transmission takes into account the transmission shaft inertia, stiffness, and damping, as well as the gear inefficiencies and torque losses due to fluid churning. Specifically, the speed reduction in each gear is assumed to be ideal, while the torque multiplication is assumed to be scaled by an efficiency factor. Furthermore, the torque lost due to fluid churning is modeled as variable nonlinear resistance of the form
FIGURE 2. THE ENGINE MODEL
2 τ chruning loss = r1 ( gear ) ωshaft + r2 ( gear ) ωshaft
(2)
where r1 and r2 are coefficients that change depending on the gear. The inputs to the shift logic, the final element in the drivetrain model, are the transmission output shaft speed and the throttle demanded by the driver. A simple chart as shown in Fig. 4 determines the current gear number and whether or not an upshift or downshift is to be initiated. The solid and dashed lines in Fig. 4
FIGURE 3. ENGINE MAP
3
Copyright © 2009 by ASME
⎛ F F2 ⎞ f rolling = sign rwheelθ& ⎜ a0 + a1 Fz + a2 z + a3 z ⎟ P P ⎠ ⎝
indicate upshift and downshift thresholds, respectively. Note that this chart is only a crude approximation of a real shift map, but it is employed here for simplicity.
(
)
(6)
where ai are empirical coefficients, and P is the tire pressure; and
(
)
τ brake = b2θ& + sign (θ& ) FCoulomb cbrake
(7)
where b2 is viscous damping coefficient, FCoulomb is the static Coulomb force, and cbrake is the brake command from the driver. This vehicle model is used in the local simulations only, where the models are connected directly or over the local area network (LAN). For the actual Internet-distributed simulations and experiments a high-fidelity multibody model of an HMMWV developed by TARDEC is employed. Internet. To characterize the Internet quality of service between UM and TARDEC, a series of experiments were run, in which packets ranging from 64 bytes to 1024 bytes were exchanged between a computer at UM and a computer at TARDEC on different days and at different times of day using the UDP/IP protocol. This protocol is preferred for its speed, but does not guarantee packet delivery. A typical result for round trip time delay vs. time of day obtained from one those experiments is shown in Fig. 5. The figure clearly shows a multi-modal character in the sense that some packets experience a delay around 25 ms, and some around 350ms (spikes), while others are dropped (shown as zero delay in the figure). A packet is considered dropped in this case if it does not arrive within 1s. Table 1 provides some statistics of the results shown in Fig. 5.
FIGURE 4. GEARSHIFT LOGIC
Client side: TARDEC. The client side includes the driver and vehicle dynamics models. The driver model is a PI controller with saturation and anti-windup. It takes the difference between the desired and actual vehicle velocities as input and generates an output within the interval [-1,1], positive values corresponding to throttle and negative values to a brake command.
The vehicle dynamics model is a point mass representation of the vehicle and includes differentials, wheel inertia, a Coulomb and viscous friction based brake model, rolling resistance, aerodynamic drag, and tire slip. The model can be expressed in the following form mvehicle && x = f slip − f aero
400
J wheelθ&& = τ differential − bθ& − τ brake − rwheel ( f rolling + f slip )
(3)
350 300
f slip
λs ⎧ λs < λs ,max ⎪ sign ( λs ) Fz μ λ =⎨ s ,max ⎪ sign ( λ ) F μ λs ≥ λs ,max s z ⎩
delay [ms]
where mvehicle is vehicle mass, J wheel is wheel inertia, b is viscous damping coefficient, rwheel is wheel radius, θ& is wheel speed, and &x& and θ&& are vehicle and wheel accelerations, respectively. Furthermore, f slip , f aero , f rolling are wheel slip, aerodynamic resistance, and rolling resistance forces, respectively, and τ differential and τ brake are differential and brake torques, respectively, and they are given by
1 Aρ air Cd x& x& 2
200 150 100 50 0
9
10
11
12
13 14 time of day [hours]
15
16
17
18
FIGURE 5. CHARACTERIZATION OF NETWORK QUALITY OF SERVICE BETWEEN UM AND TARDEC
(4)
where λs is wheel slip, Fz is the normal tire force, μ is the friction coefficient, λs ,max is the wheel slip for force saturation; f aero =
250
TABLE 1. STATISTICS FOR THE RESULTS IN FIG. 5 Number of packets 1,110,000 Packet size 1024 B Avg. delay 25.7 ms Min delay 25.3 ms No of spikes 114 (0.01%) No of drops 56 (0.005%)
(5)
where A is the vehicle frontal area, ρ air is the air density, Cd is the aerodynamic drag coefficient, and x& is vehicle velocity;
4
Copyright © 2009 by ASME
The communication architecture This work does not fully implement the event-based framework described in [9, 10], because not all signals are referenced with respect to a non-time-based event variable s. Rather, it adopts an event-based communication architecture only, primarily because of its ease of practical implementation.
0.2
occurence percentage
Actual Model 0.15
0.1
In the adopted framework the TARDEC setup acts as the client and the UM setup as the server. The client sends an updated transmission speed and throttle signal at a frequency of 20Hz, regardless of whether it receives a response or not. The server, on the other hand, only responds to the packets it receives, i.e., it only sends an updated transmission torque when it receives a packet from the client. All packets are time-stamped, and only the most recent ones are used. This ensures that both sides respond to the most up-to-date signals available.
0.05
0 0.025
0.0255
0.026
0.0265
0.027
0.0275
delay [s]
FIGURE 6. REAL VS. MODELED DISTRIBUTION OF DELAY
Based on the statistics given in Table 1, the percentage of spikes and dropped packets is very small. Thus, as a first approximation, the spikes and drops are neglected, and the emphasis is given to the characterization of the dominant mode of the delay, which appears in the darker potion of the curve in Fig. 5 around 25ms delay.
In the simulations involving the actual Internet, an increase in the network delay is simulated by bouncing the packets between TARDEC and UM multiple times before processing them. This way the delay can be increased artificially in multiples of the nominal delay. For example, if the nominal round trip delay between TARDEC and UM is approximately 25ms, an effective network delay of 125ms can be simulated by introducing 4 additional round trips. Additionally, by using the Internet itself to increase the delay, the jitter and loss statistics would scale, as well, even though this might not exactly correspond to the characteristics of a longer distance connection.
The histogram of the dominant mode of delay is shown in Fig. 6. The figure also shows that the dominant mode can be approximated quite accurately with a lognormal distribution with parameters μrt = −0.9 and σ rt = 1 , where the subscript rt indicates round trip. A scaling factor of 0.03/70 is used, along with an offset of 0.0254. Thus, the modeled distribution in Fig. 6 comes from a random variable of the form X rt ~ 0.0254 +
0.03 Log-N ( μrt = −0.9, σ rt = 1) 70
(8)
Simulation studies The model described above was utilized in a series of simulation studies. First, the driver model was given an FTP75 drive cycle to follow and the model was simulated in a single computational environment to check if the variable time delay causes instability or loss of transparency. The driver command obtained for this case is given in Fig. 7. In this case the driver response of the system was found very close to the ideal response, i.e., to the case where the coupling is direct with no time delay (also shown in Fig. 7), indicating high transparency.
The model given in Eq. (8) is for the round trip time delay. However, in the simulation a model for the one-way delay is needed. To find the parameters for the one-way delay model, first recall that for two independent random variables the mean of their sum is the sum of their means, and the variance of their sum or difference is the sum of their variances. Second, recall that the μ +σ 2 / 2 mean of a lognormal distribution is given by e , and the 2 2 variance is e 2 μ +σ (eσ − 1) . Finally, assume that the network characteristics are the same both ways. Then, the parameters μow and σ ow for the one-way delay model can be found by solving 1 μrt +σ rt2 / 2 e 2 2 2 1 − 1 = e2 μrt +σ rt eσ rt − 1 2
1
e μow +σ ow / 2 = 2
2
2
)
(
)
driver command [-]
(
e 2 μow +σ ow eσ ow
0.5
(9)
This results in the following one-way delay model X ow ~
0
-0.5
0.0254 0.03 Log-N ( μow = −1.838, σ ow = −1.221) (10) + 2 70
No delay 25ms avg delay 400ms avg delay -1
and the sum of two such one-way delays results in the desired distribution for the round trip delay.
0
50
100
150
200
250
300
350
400
time [s]
FIGURE 7. DRIVER RESPONSE FOR VARIOUS DELAYS
5
Copyright © 2009 by ASME
between UM and TARDEC to up to 100ms. The simulations were stable and did not present a perceivable transparency issue between the nominal and increased delay cases. Based on these results it was then decided to move on to the experiments with the actual hardware.
The simulation was repeated with an average delay of 400ms, which is a representative figure for the round trip delay between the East Coast of the U.S. and Australia [16]. The driver response obtained in this case is also shown in Fig. 7. Note that the driver needs to be more aggressive in this case to follow the desired drive cycle. This simulation suggests that the system has enough stability robustness, but transparency may be compromised for increased delays.
III. HARDWARE-IN-THE-LOOP EXPERIMENTS After the model-based evaluations the two hardware components, namely, the engine at UM and the ride motion simulator at TARDEC, were integrated into the simulation. The engine is a 6L V8 diesel engine with 240 kW rated power at 3300 RPM and a rated torque of 760 Nm. It is intended for a variety of medium duty truck applications covering the range between Classes IIB and VII. Because this engine is approximately twice as powerful as the standard engine used in a HMMWV, the measured engine torque was scaled by 50% before it is fed to the driveline model. Physically, this corresponds to running the engine with only half of the cylinders active. The parameters of the driveline model were also adjusted to match the known parameters for a HMMWV. A high-fidelity AC electric dynamometer coupled the physical engine with the models in real time.
The simulation was then distributed over the LAN according to Fig. 1 and repeated with the actual local network instead of the Internet model. The purpose of using the LAN was to see if the loss in accuracy in the numerical integration due to distributed simulation (with the adopted event-based communication) causes stability or transparency issues. The delay introduced by the LAN was considered negligible. The results obtained for the driver response are very similar to the ideal response. This suggests that distributed simulation does not cause major stability or transparency issues, at least for the driver response. However, in the transmission torque signal some higher frequency dynamics are excited when the simulation is distributed as seen in Fig. 8. These minor oscillations are due to the compliance of the transmission shaft. Therefore, distributed simulation causes some transparency loss in the transmission torque signal. Nevertheless, this loss is not considered to be significant, because it does not appear to affect the driver response. The effect of distributing the simulation on the transparency and the signal-dependent nature of transparency are interesting topics of discussion. This discussion is beyond the scope of this paper, but the interested reader is referred to a companion paper [28].
The ride motion simulator, on the other hand, is built on a 6DOF hydraulically-actuated Stewart platform motion base. The position, velocity, and acceleration inputs to the motion base come from the high-fidelity multi-body model of the HMMWV. The motion base together with the computer monitors installed on it provide a human operator seated on the platform with motional and visual clues, who then drives the vehicle by providing throttle, brake, and steering commands.
ideal
2500 2000
Description of experiments The stability and safety of the setup were tested experimentally with flat and hilly terrains and with accelerations and decelerations ranging from mild (ramp inputs with small slopes) to aggressive (step inputs with various amplitudes). After successful completion of these tests it was decided to move on to the full-scale experiments.
transmission torque [Nm]
1500 1000 500 0 0
50
100
150
200
250
300
350
400
LAN-distributed 2500 2000 1500
The full-scale experiments were performed with two different terrain models simulating two different closed courses at Aberdeen. The first one is the Munson Standard Fuel Course, which is a part gravel, part paved 1.67 miles long closed loop with a grade ranging from -15% to 30%. The second one is the Churchville B Course, which is an all gravel 3.7 miles long close loop with moguls and with a grade ranging from -23% to 29%. This course also contains four stop signs.
1000 500 0 0
50
100
150
200
250
300
350
400
time [s]
FIGURE 8. TRANSMISSION TORQUE FOR IDEAL AND LAN-DISTRIBUTED SIMULATIONS
Finally, the simulation was distributed over the Internet to study the effects of both distributed simulation and Internet delay. Furthermore, the point-mass vehicle model described above was replaced with a high-fidelity multibody model of a HMMWV, and the driver model was replaced with an actual driver. The vehicle was driven over various terrains, flat and hilly, and with various delay conditions, ranging from the 25ms nominal round trip delay
Four drivers were used in the experiments. Each driver drove on both courses twice; once with the nominal 25ms average delay, and once with an increased average delay of 125ms. The drivers were instructed to maintain on the Munson and Churchville courses an average speed of 20 mph and 15 mph, respectively.
6
Copyright © 2009 by ASME
vehicle speed [mph]
Experimental results Figures 9-10 show example results from one of the drivers for all experiments (Munson and Churchville with nominal and increased delay). The results are similar for the other three drivers in the sense that the experiments were stable even with the increased delay and the drivers were able to maintain the desired average speeds on both courses.
Since a direct coupling of the two hardware components is not feasible, a baseline experiment to which other experiments can be compared cannot be conducted. This prevents a rigorous transparency analysis of the experiments with the two hardware components. Nevertheless, cross-correlation can be used to obtain some comparison between the dynamics of the nominal and increased delay cases.
4
5
2
3 time [mins]
4
5
vehicle speed [mph] throttle [-]
0
2
4
6
8
10
12
14
16
8 10 time [mins]
12
14
16
throttle [-]
0.5
nominal delay increased delay sample variation in system (obtained for nominal delay)
normalized cross correlation
0.8
0.7
0.6
0
1
2
3
4
5
6
7
8
9
10
time lag [s]
FIGURE 11. SAMPLE CROSS-CORRELATIONS BETWEEN THROTTLE AND SHAFT POWER
of transparency is given in a companion paper [28]. IV. SUMMARY AND CONCLUSIONS Two hardware-in-the-loop setups, the engine-in-the-loop facility at UM and the ride motion simulator at TARDEC, are integrated over the Internet for the first time in an observer-free way. Through model-based simulation and experiments the viability of the proposed integration is demonstrated in terms of stability, robustness, and transparency. This is a first step towards identifying the fundamental limitations of an observer-free solution to Internet-distributed HILS and how to push those limits further.
6
0.5
1
6
0.4
nominal delay increased delay
0
4
0.5
1
0
2
0.9
10
3
0
1
20
2
0
FIGURE 10. SAMPLE EXPERIMENTAL RESULTS FOR CHURCHVILLE
30
1
10
0
This, however, is only a preliminary and subjective conclusion, and an objective and rigorous evaluation of transparency with a proper metric is desired. Even though an experimental analysis with the two hardware components is not feasible, analysis with using only the engine component can be performed. Such an evaluation of transparency is beyond the scope of this paper, but a more objective variation-based analysis
0
20
1
Figure 11 shows sample cross-correlations between throttle and shaft power, which can be interpreted as the demanded and delivered power, respectively. The solid and dashed curves correspond to nominal and increased delay cases of one scenario, whereas the dotted lines correspond to two different repetitions of another scenario with nominal delay and highlight the amount of variation in the experiments. In light of this variation, the crosscorrelation of the nominal and increased delay cases are considered close. According to their subjective evaluations, the drivers did not perceive any difference between the nominal and increased delay cases, either. Therefore, the transparency of the setup may be considered satisfactory for the purpose of these initial experiments.
0
30
This paper mainly outlines the steps taken to realize the Internet-distributed HILS. Experimental transparency results are preliminary, and are based on cross-correlation analysis and subjective evaluations of the drivers. A more objective evaluation is given in a companion paper [28], and further analysis is subject to ongoing research. Future work will also focus on evaluating
6
FIGURE 9. SAMPLE EXPERIMENTAL RESULTS FOR MUNSON
7
Copyright © 2009 by ASME
[13] Brudnak, M., Pozolo, M., Paul, V., Mohammad, S., Smith, W., Compere, M., Goodell, J., Holtz, D., Mortsfield, T., and Shvartsman, A., 2007, "Soldier/Harware-in-the-Loop SimulationBased Combat Vehicle Duty Cycle Measurement: Duty Cycle Experiment 2", Proceedings of Simulation Interoperability Workshop, SIW-07S-016. [14] Anderson, R. J. and Spong, M. W., 1989, "Bilateral Control of Teleoperators with Time Delay", IEEE Transactions on Automatic Control, 34(5), pp. 494-501. [15] Lee, D. and Spong, M. W., 2006, "Passive Bilateral Teleoperation with Constant Time Delay", IEEE Transactions on Robotics, 22(2), pp. 269-281. [16] Niemeyer, G. and Slotine, J.-J. E., "Toward Bilateral Internet Teleoperation", in Beyond Webcams: An Introduction to Online Robots, MIT Press, 2002, pp. 193-213. [17] Chopra, N., Spong, M. W., Hirche, S., and Buss, M., 2003, "Bilateral Teleoperation over the Internet: The Time Varying Delay Problem", Proceedings of 2003 American Control Conference, 1, pp. 155-160. [18] Munir, S. and Book, W. J., 2002, "Internet-Based Teleoperation Using Wave Variables with Prediction", IEEE/ASME Transactions on Mechatronics, 7(2), pp. 124-133. [19] Benedetti, C., Franchini, M., and Fiorini, P., 2001, "Stable Tracking in Variable Time-Delay Teleoperation", Proceedings of 2001 IEEE/RSJ International Conference on Intelligent Robots and Systems, 4, pp. 2252-2257. [20] Mirfakhrai, T. and Payandeh, S., 2005, "On Using Delay Predictors in Controlling Force-Reflecting Teleoperation over the Internet", Robotica, 23(6), pp. 809-813. [21] Kosuge, K., Murayama, H., and Takeo, K., 1996, "Bilateral Feedback Control of Telemanipulators Via Computer Network", Proceedings of 1996 IEEE/RSJ International Conference on Intelligent Robots and Systems, 3, pp. 1380-1385. [22] Hannaford, B., 1989, "Stability and Performance Tradeoffs in BiLateral Telemanipulation", Proceedings of IEEE International Conference on Robotics and Automation, pp. 1764-1767. [23] Lawrence, D. A., 1993, "Stability and Transparency in Bilateral Teleoperation", IEEE Transactions on Robotics and Automation, 9(5), pp. 624-637. [24] Griffiths, P. G., Gillespie, R. B., and Freudenberg, J. S., 2008, "A Fundamental Trade Off between Performance and Sensitivity within Haptic Rendering", IEEE Transactions on Robotics, 24(3), pp. 537548. [25] Yokokohji, Y. and Yoshikawa, T., 1994, "Bilateral Control of Master-Slave Manipulators for Ideal Kinesthetic Coupling Formulation and Experiment", IEEE Transactions on Robotics and Automation, 10(5), pp. 605-619. [26] Kazerooni, H. and Moore, C. L., 1997, "An Approach to Telerobotic Manipulations", Transactions of the ASME. Journal of Dynamic Systems, Measurement and Control, 119(3), pp. 431-8. [27] Filipi, Z. S., Fathy, H. K., Hagena, J., Knafl, A., Ahlawat, R., Liu, J., Jung, D., Assanis, D. N., Peng, H., and Stein, J. L., 2006, "Engine-in-the-Loop Testing for Evaluating Hybrid Propulsion Concepts and Transient Emissions – HMMWV Case Study", Proceedings of 2006 SAE World Congress, 2006-01-0443. [28] Ersal, T., Brudnak, M., Stein, J. L., and Fathy, H. K., 2009, "Variation-Based Transparency Analysis of an Internet-Distributed Hardware-in-the-Loop Simulation Platform for Vehicle Powertrain Systems", Proceedings of ASME Dynamic Systems and Control Conference.
different coupling points and different communication architectures for HILS integration over the Internet. ACKNOWLEDGMENTS This work was supported by a grant from the ILIR program at TARDEC to Dr. Hosam K. Fathy. The authors gratefully acknowledge this support, and would also like to thank Fernando Tavares and Rajit Johri for their help with the experiments, and Dennis N. Assanis for his feedback and support. REFERENCES [1] Fathy, H. K., Filipi, Z. S., Hagena, J., and Stein, J. L., 2006, "Review of Hardware-in-the-Loop Simulation and Its Prospects in the Automotive Area", Proceedings of SPIE - Modeling and Simulation for Military Applications, 6228, pp. 1-20. [2] Kimura, A. and Maeda, I., 1996, "Development of Engine Control System Using Real Time Simulator", Proceedings of 1996 IEEE International Symposium on Computer-Aided Control System Design, pp. 157-163. [3] Leitner, J., 2001, "A Hardware-in-the-Loop Testbed for Spacecraft Formation Flying Applications", Proceedings of 2001 IEEE Aerospace Conference, 2, pp. 615-620. [4] Ganguli, A., Deraemaeker, A., Horodinca, M., and Preumont, A., 2005, "Active Damping of Chatter in Machine Tools Demonstration with a 'Hardware-in-the-Loop' Simulator", Journal of Systems and Control Engineering, 219(5), pp. 359-369. [5] Aghili, F. and Piedboeuf, J.-C., 2002, "Contact Dynamics Emulation for Hardware-in-Loop Simulation of Robots Interacting with Environment", Proceedings of 2002 IEEE International Conference on Robotics and Automation, 1, pp. 523-529. [6] Buford, J. A., Jr., Jolly, A. C., Mobley, S. B., and Sholes, W. J., 2000, "Advancements in Hardware-in-the-Loop Simulations at the U.S. Army Aviation and Missile Command", Proceedings of SPIE Technologies for Synthetic Environments: Hardware-in-the-Loop Testing V, 4027, pp. 2-10. [7] Kelf, M. A., 2001, "Hardware-in-the-Loop Simulation for Undersea Vehicle Applications", Proceedings of SPIE - Technologies for Synthetic Environments: Hardware-in-the-Loop Testing VI, 4366, pp. 1-12. [8] Niemeyer, G. and Slotine, J.-J. E., 1991, "Stable Adaptive Teleoperation", IEEE Journal of Oceanic Engineering, 16(1), pp. 152-162. [9] Xi, N. and Tarn, T. J., 2000, "Stability Analysis of Non-Time Referenced Internet-Based Telerobotic Systems", Robotics and Autonomous Systems, 32(2), pp. 173-178. [10] Elhajj, I., Ning, X., Wai Keung, F., Yun-Hui, L., Hasegawa, Y., and Fukuda, T., 2003, "Supermedia-Enhanced Internet-Based Telerobotics", Proceedings of the IEEE, 91(3), pp. 396-421. [11] Compere, M., Goodell, J., Simon, M., Smith, W., and Brudnak, M., 2006, "Robust Control Techniques Enabling Duty Cycle Experiments Utilizing a 6-Dof Crewstation Motion Base, a Full Scale Combat Hybrid Electric Power System, and Long Distance Internet Communications", SAE Technical Paper, 2006-01-3077. [12] Goodell, J., Compere, M., Simon, M., Smith, W., Wright, R., and Brudnak, M., 2006, "Robust Control Techniques for State Tracking in the Presence of Variable Time Delays", SAE Technical Paper, 2006-01-1163.
8
Copyright © 2009 by ASME
