ANN-Based Modeling of Engine Performance from Dynamometer Sensor Data.

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Engine Performance Analysis

Background:

• Accurate prediction of internal combustion engine performance is crucial for enhancing efficiency and meeting emission standards.
• Existing methods may not fully capture the complex relationships between sensor parameters and engine load.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for predicting engine load using sensor-derived parameters.
• To demonstrate the model's accuracy and robustness in real-world automotive applications.

Main Methods:

• Utilized sensor data (design power, wheel power, torque, rotational speed) from dynamometer and hardware-in-the-loop (HiL) setups.
• Developed an artificial neural network (ANN) architecture for multiclass classification and regression tasks.
• Validated model performance using 5-fold stratified cross-validation.

Main Results:

• The ANN model achieved 99% accuracy in multiclass classification and a regression R² of approximately 0.98.
• Cross-validation yielded an average accuracy of 0.988±0.011 and a macro-F1 score of 0.984±0.011.
• The model demonstrated strong generalization and robustness in predicting complex engine load relationships.

Conclusions:

• The proposed ANN model accurately predicts engine load under normal operating conditions.
• The model's high performance and robustness make it suitable for automotive and industrial applications.
• The model is extensible for additional sensor inputs and adaptable to other powertrain systems.