Tdnn-Based Engine In-Cylinder Pressure Estimation from Shaft Velocity Spectral Representation

Area of Science:

• * Mechanical Engineering
• * Artificial Intelligence
• * Signal Processing

Background:

• * Accurate in-cylinder pressure measurement is crucial for internal combustion engine (ICE) condition monitoring and diagnostics.
• * Direct pressure sensing is invasive and complex, necessitating alternative indirect methods.
• * Shaft angular velocity is a readily measurable kinematic parameter that correlates with in-cylinder pressure.

Purpose of the Study:

• * To develop and validate a non-invasive method for estimating in-cylinder pressure in an ICE.
• * To accurately predict cylinder pressure using only shaft angular velocity fluctuations.
• * To explore the efficacy of time-delay neural networks (TDNNs) for this predictive task.

Main Methods:

• * A time-delay neural network (TDNN), conceptualized as a finite pulse response (FIR) filter, was employed.
• * The TDNN model was trained and tested using data from a single-cylinder ICE operating under 12 distinct conditions (varying angular velocity and load).
• * Model performance was optimized by adjusting the TDNN's delay parameter to maximize correlation-based scores.

Main Results:

• * The proposed TDNN methodology achieved a high prediction accuracy for in-cylinder pressure, with an R-squared (R2) value exceeding 0.9.
• * The method successfully estimated pressure from shaft angular velocity fluctuations without requiring complex pre-processing steps.
• * The TDNN's ability to capture dynamic relationships between angular velocity and pressure was demonstrated.

Conclusions:

• * Time-delay neural networks offer a robust and accurate approach for non-invasively estimating in-cylinder pressure in internal combustion engines.
• * This method provides a viable alternative to direct pressure sensing, simplifying engine condition monitoring.
• * The findings highlight the potential of kinematic data and AI for advanced engine diagnostics.