A number of approaches have been used to estimate airflow entering the cylinders in a high degree of freedom engine with variable valve timing (WT). These approaches include extra sensor information, surface look-ups, physical models, and mathematical models. For low degree of freedom engines (fixed camshaft timing), a combination of physical models and surface look-ups are the most commonly accepted method. However, as the number of degrees of freedom increases, this method becomes unwieldy because it is difficult to use surface look-ups with more than two degrees of freedom. Therefore, different approaches were studied, taking into account the following criteria: accuracy, code-ability, calibration ease, robustness across different engines, execution rate and extendability as the number of degrees of freedom increases. This paper briefly reviews previously published work that describes why Multilayer Feed Forward Artificial Neural Network (ANN) was chosen as the preferred method for approximating airflow. This paper focuses on the steps to implement the ANN method in a production embedded controller environment. These steps include creation of a fixed point model, analysis of tradeoffs between complexity/accuracy and real-time execution rates, refinement of model to code validation technique, and development of a suite of calibration tools and methods. It is shown that using an ANN for estimating airflow in an ECU is feasible. A reduced number of degrees of freedom is used to simplify the system to one ANN with four inputs and one output. The de-bugging code is accomplished by using s-functions built from the production code and testing them alongside this Simulink model. To avoid overfitting a pseudo F-statistic is proposed.


    Access

    Access via TIB

    Check availability in my library


    Export, share and cite



    Title :

    Using artificial neural networks in a production embedded controller environment for estimating airflow rate


    Contributors:


    Publication date :

    2007


    Size :

    20 Seiten, 19 Bilder, 2 Tabellen, 16 Quellen


    Type of media :

    Conference paper


    Type of material :

    Print


    Language :

    English




    VEHICLE AIRFLOW CONTROLLER

    KAWAKAMI TATSUKI / MATSUMOTO EMI | European Patent Office | 2021

    Free access