¶ 相关文件：
¶ TheNewRaceCarSimulator.doc；TetraxCornerCarRig；
¶ 1.RaceCar Simulator
(A model of flexibility)
In 1993 Servotest invent the seven post full scale race car simulator, powered by the DCS2000 controller. The car sits on four hydraulic wheel pan actuators, each capable of 25KN vertical force and +/- 125 mm displacements. These actuators reproduce the effect of the track on the tyres and the resulting forces are transmitted to the rest of the car via tyres and suspensions. The wheel pan actuators can also be used to apply sinusoidal or random inputs while a high speed data acquisition system records the car responses for subsequent analysis and modeling. In parallel three 20 kN force actuators connected to the car body apply aero-loading and weight transfer forces related to the car velocity on the track and its lateral and longitudinal accelerations. Combining this information allows Formula 1 teams to use the seven post simulator when optimising the car for each race. Similarly the race car simulator is now commonly used by F3000, Nascar, Indy or Le Mans teams.
10 years later, Servotest Systems launch their latest platform, Lumina, successor of the DCS2000 digital controller. Lumina is an external real time DSP controller that uses FireWire technology (IEEE 1394a) for easy connection to a desktop PC or a laptop. Extra computing power is provided by intelligent distributed hub-node architecture using proven fiber optic link technology. For maximum portability the front end was developed using object oriented C# within the .NET Framework. Its event driven architecture promotes both flexibility and reliability.
Model in the loop simulation
The race car simulation software (see Figure 1) has been entirely re-written to make the most of the extra power and flexibility provided by Lumina and to meet the ever growing needs of test and development engineers. In particular, over the past decade the importance of model-based simulation has grown remarkably. A typical example is that of the tyre model. The characteristics of a rotating tyre are very different to those of a static tyre. Therefore, provided the rotating tyre can be accurately modeled, it becomes advantageous to replace the true static tyre of the car on the simulator with a virtual tyre whose characteristics vary with the car velocity and other parameters. This takes the form of a load vs. displacement DSP model whose output force is applied straight to the hub of the spindle-coupled car. The DSP model (also known as Socket) is actually produced from a Simulink representation of the dynamics using the Real Time Workshop and a DSP compiler. The whole process of producing a Socket based on a Simulink model and plugging the Socket into the controller DSP is expedited by the Socket Wizard.
Figure 1: The new race car simulator front end
In the Lumina environment each and every DSP object can be connected to a Socket. In other words, the test engineer has full control over the control laws and the dynamics that are implemented in the DSP. For example, although it is common to drive the down force based on the car velocity and the front and rear ride heights, it is straight forward to add damping factors that would apply to the rate of change of the ride heights. Similarly, if it is judged that the banking force plays a significant role in the behaviour of the car (for example on oval tracks) vertical acceleration can easily be incorporated in the calculation of the aero-loader commands (see Figure 2). The Simulink model can take the shape of a block diagram including linear or non-linear elements as well as multi-dimensional look-up tables as readily available in Simulink. Alternatively, more complex models can be represented by C based S-functions. Using C language to program sockets has recently been made easier by the appearance of Simulink’s S-function Builder. This guides the user through the specification of all the parameters and methods that are necessary to the correct execution of C S-functions.
Figure 2: Implementing aero damping and banking force using a Socket
The emergence of new linear identification tools such as Matlab’s System Identification Toolbox as well as non-linear modeling techniques such as neural networks means that we are more and more capable of producing mathematical models that accurately represent the static and dynamic behaviour of any system. Having said that, when we know that theoretically an estimated 150 states are necessary to model the three dimensional dynamics of a tyre, it is reasonable to assume that the modeling cycle (including model order reduction and validation) is likely to remain the bottle neck of the model-based simulation exercise. Still, it is reassuring to know that with tools such as Lumina’s Socket Wizard, the introduction in the loop of the long awaited model becomes a non-event.
Total flexibility
Thanks to the distributed architecture of Lumina and its front end, the race car simulator is not so much a seven post simulator as a simulator with seven posters. In its basic configuration the software allows each actuator to be controlled independently of the others. This means that any of the seven actuators can be used to apply stimuli to the car either through the tyres or the body. The effect of these local inputs on other parts of the car can then be measured, analysed and modeled.
In other cases, it is important to be able to link some of these actuators together in order to facilitate the use of the simulator, or to make it safer. For example in the case of a spindle-coupled car, applying a large downwards displacement to a single of the four corners by accident could result in extensive damage to the suspension. However, the same displacement applied to all four corners simultaneously would be perfectly safe. So the software lets the user build their own configurations, adapt them to specific tests and change the aspect of the front end to reflect these needs. The software is thereby infinitely expandable and uniquely matched to the needs of the user.
The following situation illustrates the need for a flexible and configurable platform. The aero-loaders are typically positioned in a symmetrical triangular fashion so that the left and right actuators are used to apply cornering (roll) moment and, combined with the centre aero-loader, to apply breaking (pitch) moment. In the case of Nascar racing, this arrangement is particularly suited to the simulation of circuits with both left and right turns and necessitating repeated breaking and acceleration. However, in the case of ovals, it can be argued that the pitching moment and the dynamic part of the rolling moment play a small role compared to the other forces (particularly centrifugal) that produce asymmetrical loading on the car. In such a situation spreading the load may be beneficial and therefore being able to easily move the aero-loaders to have two on the “heavier” side of the car and one on the other is appreciable.
In Lumina, the calibration and linearisation information of all sensors and valves related to an actuator is stored in a local DSP module that remains with the actuator. This makes all the actuators completely stand-alone and interchangeable. This notion is reinforced by the use of light fiber optic cables for data transmission instead of the traditional copper cables. This distributed arrangement also means more DSP computing power is available to implement real time model simulation in the master DSP controller.
For too long engineers have had to adapt their tests to the equipment at their disposal and work around its limitations. Hopefully the time has come when the hardware components of a test setup (such as actuators and transducers) can truly be seen as individual blocks that the test engineer can drag and drop anywhere on his flow chart to best suit the objective of the test he is setting up.
Test and Simulation Suite applied to curb testing
A classical problem facing racing engineers is the behaviour of the car as it goes over curbs at high speed. On one hand the suspension setup must be optimised to allow the car to go over curbs as quickly as possible, on the other hand the engineers must check that the huge accelerations and forces generated will not compromise the integrity of car after 50 or 80 laps. This simulation problem can be tackled in two main steps: replication and endurance.
Figure 3: Test and Simulation Suite
The Iteration Control System (ICS) is a replication package that iteratively generates the drives that will allow certain desired responses to be accurately reproduced by the rig. These desired responses could be either pre-recorded time histories (e.g., the car hub accelerations as it goes round the track) or, as pictured in Figure 3, artificial channels produced by a desired file generator (e.g., some random signals whose PSD follows a user defined profile). In the case of a curb, the desired file would typically be a seven channel artificial file whose first four channels represent the displacement of the four wheels as the car goes over the curb. The last three channels are load commands sent to the three aero-loaders to represent the loading of the car at this instant (i.e., down force, cornering force, etc…). ICS (pictured in Figure 4) is used to minimise the error between the desired and actual responses of all seven channels. This is done by first computing a linear model of the input-output relationship using a Matlab based Maths Toolbox (see Figure 3), and then using this system matrix to compute the successive drives. Because each actuator is controlled individually, very few iterations are required to obtain the drive file that accurately simulates the curb.
Figure 4: The Inductive User Interface for ICS
Using this drive file the test engineer can now very easily see the effect of various suspension characteristics on the car behaviour and therefore attempt to find the optimum setup for this condition.
Subsequently, the drive file is used as part of an endurance test sequence that attempts to identify weak links in the overall system structure. Lumina’s test sequencer called EZFlow (see Figure 5) is used to program the sequence that will see the drive file played over and over again until either the total number of laps is reached or the system detects signs of fatigue or failure. These checks take place either instantaneously or on a file by file basis using Lumina’s Trend Monitor. The successive response files are automatically scanned and various statistics are computed and compared with user defined limits. If any of these limits is exceeded the Trend Monitor decision block executes the action specified by the user within EZFlow. The responses can also be analysed both in the time domain and the frequency domain using a Data Analysis Package and a Real Time Display (e.g., spectrum analyser) provided by the Test and Simulation Suite.
Figure 5: EZFlow test sequencer
¶ 2. Tetrax Corner Car Rig
At the time of writing this article, Servotest Systems have installed 2 Texraxial Test rigs, one at AMG in Germany and one at Redbull Racing F1.
The key issue with a Tetraxial rig is how the geometry of the linkages is correctly translated and communicated to the test engineer.
Geometry
The Longditudinal and Vertical displacment signals are factored by multiplying the Scale of the Signal Conditioner objects by the lever ratios of the Y-shaped strut and the bell crank accordingly. The Bell crank for example is a 3:1 step-up ratio so the working stroke of the actuator of 25mm is multiplied by 3 to give the Vertical displacment scale of 75mm.
The Vertical actuator Load and Displacment signals are inverted to reflect the Bell-Crank inversion, i.e., extension of the actuator results in downward motion of the wheel-hub, which in turn is negative wheel-hub displacment and positive suspension load.
There is also an elaborate system to correct for geometry using a system of routers and a specially written piece of software. In retrospect, this system has proved to be too much effort and these problems are solved better by ICS.
Mass Cancellation
There is mass cancellation on the Longditudinal Load, the Lateral Load and Braking Load but NOT on the vertical load. It was deemed that mass cancellation as secondary to damping oscillations with Delta-P.
Control
The PID terms are deliberately de-tuned for load control due to the fact that the specimen stiffness is part of the loop gain and sudden changes in gain can have destabalising effects.
The directions of positive Force and Displacement are the same in the Longditudinal, Lateral and Braking axes but opposed in the Vertical axis. This means that to consistently get negative feedback, the F/B scale for the load servocontroller is negative. For some reason, the vertical load cell is guaged so that Compression is positive and tension is negative, hence the negative X-Terms to give compression as negative (in order to be in line with automotive convention).
In addition, the displacment X-Terms have been inverted so that actuator extension results in negative wheel hub motion. For this to work, it appears the servovalve has been mounted upside down
The side-effect of this is that the actuator response in displacement is opposite to the command signal. It is therefore necessary to use -100% as the input scales for the load controller.