A Cylinder Pressure Recovery Method for Diesel Engine Based on GMM Algorithm and BP Neural Network

• Published in: Chinese Control Conference pp. 3804 - 3808
• Main Authors: Sun, Yubo, Su, Liwang, Guan, Zhuowei, Xiao, Wei, Yang, Chenfan, Bai, Wei, Dong, Jiangfeng, Chu, Quanhong
• Format: Conference Proceeding
• Language: English
• Published: Technical Committee on Control Theory, Chinese Association of Automation 28.07.2025
• Subjects: BP Neural Network; Cylinder Pressure Recover; Diesel Engine; Engines; Gaussian Mixture Modeil; Heuristic algorithms; Intelligent Power System; Neural networks; Power system dynamics; Prediction algorithms; Real-time systems; Signal processing; Signal processing algorithms; Vehicle dynamics; Velocity control
• ISSN: 1934-1768
• DOI: 10.23919/CCC64809.2025.11178538

The demand for cylinder pressure signals in closed-loop control and health diagnosis during the operation of automotive engines is addressed in this study. In response to the challenges associated with the difficult and cumbersome acquisition of cylinder pressure signals, an innovative methodology for parameterizing and predicting cylinder pressure curves in internal combustion engines is proposed. This methodology integrates Dynamic Time Warping (DTW), Gaussian Mixture Model (GMM), and a Backpropagation (BP) neural network. Characteristic curves and parameter arrays are extracted from cylinder pressure signals using the DTW and GMM algorithms, providing a robust representation of pressure dynamics. The BP neural network is subsequently trained with engine speed and acceleration signals as inputs, while the extracted pressure parameters are utilized as outputs, enabling the prediction of cylinder pressure based on rotational speed data. This approach eliminates the necessity for direct pressure measurements, offering a more practical and efficient solution for engine diagnostics and performance analysis. The effectiveness of the proposed method in accurately recovering cylinder pressure curves is demonstrated through the results, highlighting its potential for applications in combustion analysis, engine optimization, and real-time monitoring systems. A bench test is designed to validate the feasibility and effectiveness of the proposed method. By combining advanced signal processing techniques with machine learning, this study contributes to addressing challenges in engine pressure modeling and prediction, thereby advancing the field.