Automotive Digest - November 2003

DENSO uses the Model-Based Calibration Toolbox in the Development of Automotive Powertrain Systems

By: Manji Suzuki and Masahiko Kimura
DENSO Corporation

DENSO CORPORATION, a global automotive supplier of advanced technology, systems, and components, has been using The MathWorks Model-Based Calibration Toolbox for the past 22 months. DENSO uses the toolbox to calibrate its engine control units for new engines that have more control parameters to take into account than previous designs. This leads to a significant increase in the calibration effort, which actually needs to be reduced in order to lower the cost and time to market.

The calibration effort rapidly expands as the number of control parameters increases. For previous engine designs, there were just a few, but the latest engines can have up to seven or eight additional control parameters including, for example, injection timing, exhaust gas re-circulation rate (EGR) and variable cam timing (VCT). The MathWorks has been assisting DENSO in developing its new strategy for calibrating engines, including building a mathematical model of a prototype engine, understanding what’s going on in the engine and then calibrating it using the model. This is all done using the Model-Based Calibration Toolbox.

DENSO has been applying two methods simultaneously to reduce calibration efforts:

i. The general concept of model-based design. i.e. front-loading of calibration efforts by using plant models and a HIL (hardware-in-the-loop) system. This means undertaking as much simulation as possible ahead of time rather than doing all the testing on a real engine. This will increase the speed of the development process and reduce the number of expensive prototypes that need to be built.
ii. Introducing automatic map (or look-up table) generation techniques by using the Model-Based Calibration Toolbox. i.e. automatically generating the required calibrations.

Conventionally, the calibration of an engine is mainly done on the engine test bed or in-vehicle using the real engine control unit (ECU) and the real engine. DENSO is working to shorten the development cycle by front-loading calibration in the early phases of design using models of the engine and vehicle. These models are developed from test data obtained from an initial prototype of the engine. They can then be reused in all areas of the design process. At present, the company is doing offline and online virtual calibration at the research stage.

To calibrate a map conventionally, the engine is measured at a grid of data points and the measured data is entered directly into the map (rather than taking the data points and fitting a model to them). Using the Model-Based Calibration Toolbox, DENSO is selecting experiments to run on the engines with new types of design-of-experiment (DoE), such as the space filling method. With this, data can be collected from the engine on the test bed (using an automated operation and measuring system) and a model can be built from that data. The look-up tables that form the maps can then be filled using the models. Previously, simple linear interpolation was performed on the data points; with this new process more sophisticated models, for example, radial basis function (RBF) models are used, which generate more accurate calibrations.

The Model-Based Calibration Toolbox consists of the Model Browser and CAGE (Calibration Generation). DoE and modeling can be achieved with the Model Browser, and the mapping process can be performed with CAGE. DoE involves working out the most intelligent places to collect data points. The design can be created, constrained, and evaluated before it is exported. Modeling involves creating the models, removing bad data points, evaluating the model, selecting the best model type, and exporting the model. The outputs from the DoE and the resulting models are then used for CAGE, which converts an n-D map into a combination of 1-D/2-D maps (known as a feature strategy). The breakpoints (location of the grid points in the maps) and values of the feature strategy can be optimised to fit the model with CAGE.

An example of how the Model-Based Calibration Toolbox is being used by DENSO is a standard application for engine characteristic map calibration: modeling airflow as a function of manifold pressure and other variables, such as engine speed, engine load and EGR. Here, the target map is the ‘airflow-manifold pressure strategy’ that is included in a typical torque base engine control strategy. In this strategy, airflow is the input and manifold pressure is the output, and what needs to be understood is how the behaviour of this relationship is influenced by VCT, EGR and engine speed.

If a model isn’t used, the airflow-manifold pressure relationship is measured at a number of points in each of the other variable directions. i.e. at a large grid of points. This is very expensive because of the sheer amount of data that is measured. Many tables are required for expressing the behaviour and it takes a long time to measure the huge data sets for the multiple tables. In addition, a large amount of costly storage space is required on the ECU.

The alternative is to build a model of manifold pressure from a smaller set of data points and then to use that model to understand its behaviour at all the other input points. This reduces the number of data points measured, which reduces cost and time to market. The model can be validated by taking more measurements from the engine and seeing how well the model predicts the data at those points. By modeling the relationship the inherent error that comes from measuring the data is removed. The accuracy of the model depends on the number of data points collected and whether it is flexible enough to pick up all the trends in the data without being adversely affected by the noise (error).

Rather than just selecting any set of data points to be measured, a set of points is chosen that will provide the most useful information from which to build an accurate model. Space filling DOE provides a good scatter of data points in the space to be measured. The system is then modeled by trying to understand the trends in the data. DENSO modeled manifold pressure as a function of the five input variables using a RBF (polynomials can be too simple to describe complex responses, especially in a high number of dimensions, so a more sophisticated function is needed). This is just one model type offered by the Model-Based Calibration Toolbox.

Using the Model-Based Calibration Toolbox, the model can be viewed and evaluated in a number of ways, so that its accuracy can be assessed or a detailed analysis or investigation of its robustness can be carried out. It is then validated with data from the engine – this is a very important step.

Lastly, the model can be used to develop a strategy that will be used in the ECU. This is done using CAGE. A strategy can be developed in a clever way so that it takes into account the structure of the model. The feature strategy developed by DENSO for the airflow-manifold pressure feature required around ten tables to capture the model behaviour (compared to the hundreds or thousands required from traditional techniques).

The MathWorks is currently working with DENSO to develop the Model-Based Calibration Toolbox by extending particular features of the toolbox to customize it to the company’s processes. These are provided to DENSO as add-ons to the existing functionality.

---

To Subscribe to upcoming issues of The MathWorks Automotive Digest, please send an email to digest@mathworks.com