4569846498

01.05.2017 Views

Modelling and assembly of the full vehicle 373 equally it can obscure performance changes and so adaptive modelling of driver behaviour is not preferred except for circuit driving. Several techniques come under the headline of adaptive control; the simplest is to change the control parameters in a predetermined fashion according to the operating regime, an operation referred to as ‘gain scheduling’. Gain is the term used for any treatment given to an error state before it is fed to an input – thus the PID controller described above has a P-gain, an I-gain and a D-gain. It might be, for example, that under conditions of opposite lock the P-gain is increased since the driver needs to work quickly to retain control, or under conditions of increasing speed the P-gain is reduced since slower inputs are good for stability at higher speeds. A more complex method is to carry a model of the plant on board in the controller and to use it to better inform some form of gain scheduling, perhaps using information that cannot readily be discerned from on-board instrumentation – such as body slip angle. This is referred to as a Model Reference Adaptive Scheme (MRAS). A further variation on the theme is to use the controller to calculate model parameters using system or parameter identification methods (described above). The control system parameters can be modified based on this information – in effect there is an ongoing redesign of the control system using a classical deterministic method, based on the reference state and the plant characteristics according to the latest estimate. This is referred to as a ‘self-tuning-regulator’ and is useful for unpredictably varying systems. Finally, a method known as ‘dual control’ intentionally disturbs the system in order to learn its characteristics, while simultaneously controlling it towards a reference state. In many ways this is similar to a top level rally driver stabbing the brakes in order to assess friction levels while disturbing the overall speed of the vehicle as little as possible; the knowledge gained allows the driver to tune their braking behaviour according to recently learned characteristics. Such behaviour is in marked contrast to circuit drivers, who concentrate on learned braking points and sometimes have difficulty adapting to changing weather conditions. With the exception of the simplest gain scheduling methods, in general adaptive control techniques are unsuitable for the modelling of driver behaviour as part of any practicable process. Once again the variation in simulation output cannot readily be traced to any particular aspect of the system and hence the success or otherwise of an intended modification is difficult to interpret. In the light of the preceding description, the authors believe a PID controller, with some form of simple gain scheduling, is most appropriate for the modelling of driver behaviour in a multibody system context. The art of implementing a successful model is in selecting the state variables within the model to use with the controller. 6.13.2 A path following controller model The first hurdle to be crossed is the availability of suitable state variables and the use of gain terms to apply to them. Typically in a multibody system model, many more variables are available than in a real vehicle. Within the model, these variables can be the subject of differential equations in order to have available integral and differential terms. Table 6.4 shows a portion

374 Multibody Systems Approach to Vehicle Dynamics Table 6.4 A portion of an MSC.ADAMS command file showing the implementation of differential equations to retrieve and use integral and derivative terms for a state variable ! -- Derivative Term - not generally used -- part create equation differential_equation & differential_equation_name .test.yaw_rate_error_equation_1 & adams_id 3 & comments “Yaw Rate Error Equation - Implicit” & initial_condition 0.0 & function “DIF(3)-varval(yaw_rate_error)” & implicit on & static_hold off data_element create variable & variable_name yaw_rate_error_derivative & function “DIF1(3)” ! -- Integral Term -- part create equation differential_equation & differential_equation_name .test.yaw_rate_error_equation_2 & adams_id 4 & comments “Yaw Rate Error Equation - Explicit” & initial_condition 0.0 & function “varval(yaw_rate_error)” & implicit off & static_hold off data_element create variable & variable_name yaw_rate_error_integral & function “(DIF(4))” ! Steer input torque in response to path error. force create direct single_component_force & single_component_force_name yaw_rate_handwheel_torque & type_of_freedom rotational & action_only on & i_marker_name .hand_wheel_column.m_wheel_column & j_marker_name .hand_wheel_column.m_wheel_column & function “(“, & “ VARVAL(yaw_rate_error) * VARVAL(yp_gain) ”, & “ VARVAL(yaw_rate_error_integral) * VARVAL(yi_gain) ”, & “ VARVAL(yaw_rate_error_derivative)* VARVAL(yd_gain) ”, & “)”, & “* STEP(TIME,0.0,0.0,1.0,1.0)” of a command file from MSC.ADAMS implementing those terms for yaw rate. While it is a working example, no claim is made that is in any sense optimum. Such variables can usually be manipulated within the model using the programming syntax provided with the code being used. For simulation codes such as MSC.ADAMS, the format of such calculations can appear a little clumsy but this soon disappears with familiarity. The most recent versions of MSC.ADAMS include a ‘control toolbox’ to facilitate the implementation of PID controllers. For codes such as MATLAB/Simulink the implementation of control systems is arguably easier since they are written with the prime objective of control system modelling. However, the modelling of the vehicle as a plant is more difficult within these systems and so there

Modelling and assembly of the full vehicle 373<br />

equally it can obscure performance changes and so adaptive modelling of<br />

driver behaviour is not preferred except for circuit driving. Several techniques<br />

come under the headline of adaptive control; the simplest is to<br />

change the control parameters in a predetermined fashion according to the<br />

operating regime, an operation referred to as ‘gain scheduling’. Gain is the<br />

term used for any treatment given to an error state before it is fed to an<br />

input – thus the PID controller described above has a P-gain, an I-gain and<br />

a D-gain. It might be, for example, that under conditions of opposite lock<br />

the P-gain is increased since the driver needs to work quickly to retain control,<br />

or under conditions of increasing speed the P-gain is reduced since<br />

slower inputs are good for stability at higher speeds. A more complex<br />

method is to carry a model of the plant on board in the controller and to use<br />

it to better inform some form of gain scheduling, perhaps using information<br />

that cannot readily be discerned from on-board instrumentation – such<br />

as body slip angle. This is referred to as a Model Reference Adaptive<br />

Scheme (MRAS). A further variation on the theme is to use the controller<br />

to calculate model parameters using system or parameter identification<br />

methods (described above). The control system parameters can be modified<br />

based on this information – in effect there is an ongoing redesign of the<br />

control system using a classical deterministic method, based on the reference<br />

state and the plant characteristics according to the latest estimate. This<br />

is referred to as a ‘self-tuning-regulator’ and is useful for unpredictably<br />

varying systems. Finally, a method known as ‘dual control’ intentionally<br />

disturbs the system in order to learn its characteristics, while simultaneously<br />

controlling it towards a reference state. In many ways this is similar<br />

to a top level rally driver stabbing the brakes in order to assess friction levels<br />

while disturbing the overall speed of the vehicle as little as possible; the<br />

knowledge gained allows the driver to tune their braking behaviour according<br />

to recently learned characteristics. Such behaviour is in marked contrast<br />

to circuit drivers, who concentrate on learned braking points and<br />

sometimes have difficulty adapting to changing weather conditions. With<br />

the exception of the simplest gain scheduling methods, in general adaptive<br />

control techniques are unsuitable for the modelling of driver behaviour as<br />

part of any practicable process. Once again the variation in simulation output<br />

cannot readily be traced to any particular aspect of the system and<br />

hence the success or otherwise of an intended modification is difficult to<br />

interpret.<br />

In the light of the preceding description, the authors believe a PID controller,<br />

with some form of simple gain scheduling, is most appropriate for<br />

the modelling of driver behaviour in a multibody system context. The art of<br />

implementing a successful model is in selecting the state variables within<br />

the model to use with the controller.<br />

6.13.2 A path following controller model<br />

The first hurdle to be crossed is the availability of suitable state variables<br />

and the use of gain terms to apply to them. Typically in a multibody system<br />

model, many more variables are available than in a real vehicle. Within the<br />

model, these variables can be the subject of differential equations in order<br />

to have available integral and differential terms. Table 6.4 shows a portion

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!