Ca50 estimation based on Neural Network and smooth variable structure filter.

Area of Science:

• Internal combustion engine combustion analysis
• Thermodynamics and heat transfer in engines
• Sensor technology and signal processing

Background:

• CA50 and indicated mean effective pressure (IMEP) are critical for engine combustion analysis and performance evaluation.
• In-cylinder pressure sensors are essential for calculating CA50 and IMEP, but sensor degradation or failure compromises accuracy.
• Measurement errors, particularly pressure offset and thermal shock, significantly affect the reliability of these key engine parameters.

Purpose of the Study:

• To analyze the impact of various measurement error sources on CA50 and IMEP calculations.
• To develop a robust method for improving the accuracy and reliability of CA50 calculation despite sensor inaccuracies.
• To validate the proposed method using experimental data.

Main Methods:

• Analysis of the effects of pressure offset and thermal shock on CA50 and IMEP.
• Development of a neural network model to predict CA50 using in-cylinder parameters (temperature, pressure, O2 concentration, injection mass/time).
• Integration of pressure-derived CA50 with the neural network-predicted CA50 using a smooth variable structure filter.

Main Results:

• Thermal shock was found to reduce IMEP calculation accuracy.
• CA50 demonstrated sensitivity to all analyzed measurement errors.
• The proposed neural network and filter method successfully improved the accuracy and robustness of CA50 calculation.

Conclusions:

• Sensor errors significantly degrade the accuracy of critical combustion parameters like CA50 and IMEP.
• A hybrid approach combining neural networks and filtering techniques offers a reliable solution for accurate CA50 estimation.
• The validated method enhances the dependability of engine combustion monitoring systems.