Electronic Throttle Control (ETC) is a drive-by-wire technology ubiquitous in modern vehicles. Both diesel and gasoline engines have greatly benefited from this technology, with considerable improvements in terms of fuel consumption, pollutant emissions and vehicle drivability. As the Engine Management System (EMS) of each and every modern vehicle relies heavily on the performance of this servomechanism, the underlying control system must be efficient, robust and easily tunable.
The principle of ETC is a simple one, since it comes down to controlling a low-power DC servo drive in order to position the throttle plate. However, a number of issues make application of the principle a little more difficult, especially in a mass-production context: friction, limp-home nonlinearity and parameter dispersion.
Since we are convinced that this control problem is of both industrial and academic interest, we are offering the automotive control community a benchmark to evaluate throttle control systems, designed and calibrated following different approaches, in frameworks of increasing complexity. A “black box” Simulink model, together with a series of test cases and control specifications, will be provided to let participants in the benchmark design their control system.
The resulting control systems, coded in a real-time compatible Simulink block will undergo a sequence of validation tests:
first of all, in co-simulation with the nominal control model (with a wider set of test cases);
then, in a Software-in-the-Loop (SiL) environment with real-time constraints;
finally, in a Hardware-in-the-Loop (HiL) environment, which includes a real throttle.
For the real-time evaluation part, we invite the participants to provide a procedure to recalibrate theirs controllers on the real throttle. Participants in the ETC benchmark will be asked to submit a paper describing their approach and their results on the nominal model. Their papers will be included in a special E-COSM’09 session.
Benchmark description
Part 1: Simulation
A throttle model is given in the form of a Simulink S-function corresponding to a nonlinear model whose parameter have been identified from experiments on a real throttle with limp-home. The model is accompanied by a sample control system (a sluggish PID) to be replaced with a control system proposed by the participants.
The input to the model is a saturated PWM signal, and its output is throttle position expressed in % (indicated as %pos in the following, with 100%pos corresponding to fully open throttle and 0%pos to fully closed throttle).
A first set of reference signals (A) is provided together with the throttle model:
A1. Steps and combinations of steps (90%pos, 50%pos, 10%pos); A2. Ramps; A3. More complex signals (cycles).
The primary task of the benchmark consists in designing a controller satisfying a set of specifications with the given throttle model. Specifications for control design are given as follows:
with signals A1
- static error should not be greater than 0.125%pos (quantization error);
- settling time (at ±5%) should not be greater than 200ms for amplitudes over 50%pos and than 100ms for smaller amplitudes;
- maximum overshoot should not be greater than 0.5%pos for amplitudes over 50%pos and than 0.125%pos for smaller amplitudes;
with signals A2, tracking error should not be greater than 2.5% of ramp slope (expressed in %pos/s);
integral square error (ISE) must be minimized for all signals (A1, A2, A3).
The controller must be provided as a Simulink subsystem to be executed at sample times larger than or equal to 1ms and multiple of the base sample time of the nominal model S-function (200µs).
The performance of proposed controllers will be first assessed using the nominal model and the nominal reference signals. A second set of reference signals (B), not provided to participants, will be used for further assessment.
Part 2: Real-time
The second part of the benchmark consists in evaluating real-time control performance on a real throttle, whose nominal model has been given in part 1. Preliminary tests in a software-in-the-Loop (SiL) environment including the nominal model will helpd find out whether the proposed controller is compatible with real-time constraints. Finally, the proposed controller will be tested, in Hardware-in-the-loop environment, with the reference signal sets A and B.
Since the nominal model captures only partially the actual dynamic behavior of the system, we recommend the participants to provide in addition a procedure to recalibrate their controllers from experiments on the real throttle. Whilst not compulsory, the design of such a procedure would be sensible practice and add increase industrial relevance of the proposed approach. Contributors should provide a calibration code (.m or .p file) together with a set of input signals (.mat file). A experiment with these signals will be run on the real throttle and the results will be used as an input to the calibration function. The calibration function will calculate the new parameters to be used in the controllers (.mat file).
The same evaluation as with the nominal controller will be carried out with the recalibrated controller.
How to participate
All the files needed to start working on the Throttle Control Benchmark are contained in the archive file TC_benchmark.zip.
If you are interested in participating in the benchmark, please contact the organizers Gianluca Zito (
) or Paolino Tona (
). Your name will be added to a mailing list to keep you informed of developments and updates.
All the benchmark deliverables, which include the files used to implement and initialize the nominal controller, and the calibration procedure (if submitted), are due by 15 July, 2009. By the same time, participants are requested to submit a paper describing their approach and their results on the nominal model at IFAC Papercept: http://ifac.papercept.net/
