Developing a Technique for Extraction of Transient Control Strategies from Automotive Electronic Control Units, 03-9247

Printer Friendly Version

Principal Investigators
Yiqun Huang
Arun Vemuri
Jayant Sarlashkar

Inclusive Dates: 04/01/01 - 04/01/03

Background - Automotive system benchmarking is a common practice used by both automotive companies and government agencies. The benchmarking process typically involves the characterization of the mechanical as well as the electrical aspects of the automotive system. This process of characterization is used not only to evaluate automotive products that outperform their competition but also, more importantly, to create a "baseline." This baseline, which is a system equivalent in characteristics to the automotive product, is often used to develop and add new features to the product, thus enhancing the product.

The automotive control system comprises of a number of sub-control systems such as the fuel quantity control system, fuel timing control system, and EGR control system. Each of these sub-control systems typically consists of two components: (1) the steady-state component and (2) the transient component.

The steady-state component, as the name implies, is responsible for the control of the relevant subsystem during steady-state operation while the transient component is responsible for controlling the sub-system behavior during transient events such as a throttle tip-in and tip-off.

This project investigated the characterization process of the transient-component of a production automotive engine control system - fueling control system.

Approach - This project developed a general methodology for characterizing the automotive engine controller based on the investigation of the characterization process for a specific engine. The control strategy benchmarking process included the extraction of both the steady-state and the transient control strategies. The steady-state component of the control was identified using both the bilinear interpolation technique and neural network model. For the transient component, four different model structures were examined.

Accomplishments -

  1. This project demonstrated that the steady-state behavior of an engine control unit can be accurately reproduced using a neural network based model. The neural network model has to be trained with sufficient input-output data from the controller. The inputs to the model and model structure should represent the real world controller as closer as possible in order to achieve higher accuracy.
  2. The project also demonstrated that the transient behavior of an engine control unit can be predicted using different neural network based models, such as PI based neural network model with RPM and Pedal position as inputs, dual static neural network models with DRPM and DPedal Position as inputs and dual neural network models with DRPM and DPedal Position and fuel pressure as inputs.
  3. Because of the multi input variable nature of the real-world engine controller, the transient approximators can not reproduce all transient controls from the OEM ECU by using rpm and pedal position inputs. To improve the accuracy of the model, more understanding of the input parameters to the production control system is necessary.
Figure 1. Estimated Rail Pressure Regulator Duty Cycle Map Using the Neural Network Technique Figure 2. Estimation of Rail Pressure Regulator Transient Duty Cycle from the Estimator with DRPM, DPedal and DPressure as Inputs to the Transient Nueral Networks

2003 Program Home