Scholarlink

<p>This study implements machine learning approaches, specifically Artificial Neural Networks (ANN) and transfer learning, to predict and optimize the operating conditions of pilot and main injections in a common rail diesel fuel injection system, based on input parameters including injection pressure, injection frequency, and pulse time. The results demonstrate that the ANN model achieves high predictive accuracy (R2 = 0.9885, RMSE = 0.0366 ,and MAE = 0.0232). Moreover, applying transfer learning to the pilot injection dataset further improves prediction performance (R2 = 0.9931, RMSE = 0.0274, MAE = 0.0240). Data analysis identifies the optimal operating regions as follows: at low injection frequencies (500–1000 cycles/min), the injection rate remains low and stable under high pressures (1600–2400&#xa0;bar). At medium injection frequencies (1100–2000 cycles/min), the injection rate increases significantly, showing high sensitivity to variations in pressure and pulse duration, with local peaks observed at 1400–1700&#xa0;bar. At high injection frequencies (2000–2600 cycles/min), the pilot injection rate slightly decreases while the main injection rate increases, exhibiting large variations due to short pulse durations, though greater stability can be achieved when combining 1400–2000&#xa0;bar with longer pulse durations. Overall, injection pressure plays the dominant role, whereas pulse duration and injection frequency determine the effective needle opening time and the sensitivity of the injection rate. Selecting the optimal parameter combinations for each frequency range ensures stable and repeatable fuel injection rates that meet operational requirements. Consequently, this study provides a scientific basis for subsequent research on machine-learning-based optimal control strategies for dual-fuel diesel injection systems, aiming to reduce emissions, enhance fuel economy, and improve engine performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Experimental and transfer learning - machine learning approaches for predicting optimal operating conditions of pilot and main injection in common rail diesel injectors based on injection pressure and pulse duration

  • 期刊论文

  • 发表日期：

    2026年8月3日

查看全文
  • Nguyen Xuan Khoa,
  • Ta Duc Quyet,
  • Nguyen Xuan Hien,
  • Hoang Quang Tuan,
  • Vu Hai Quan,
  • Nguyen Phi Truong

摘要

This study implements machine learning approaches, specifically Artificial Neural Networks (ANN) and transfer learning, to predict and optimize the operating conditions of pilot and main injections in a common rail diesel fuel injection system, based on input parameters including injection pressure, injection frequency, and pulse time. The results demonstrate that the ANN model achieves high predictive accuracy (R2 = 0.9885, RMSE = 0.0366 ,and MAE = 0.0232). Moreover, applying transfer learning to the pilot injection dataset further improves prediction performance (R2 = 0.9931, RMSE = 0.0274, MAE = 0.0240). Data analysis identifies the optimal operating regions as follows: at low injection frequencies (500–1000 cycles/min), the injection rate remains low and stable under high pressures (1600–2400 bar). At medium injection frequencies (1100–2000 cycles/min), the injection rate increases significantly, showing high sensitivity to variations in pressure and pulse duration, with local peaks observed at 1400–1700 bar. At high injection frequencies (2000–2600 cycles/min), the pilot injection rate slightly decreases while the main injection rate increases, exhibiting large variations due to short pulse durations, though greater stability can be achieved when combining 1400–2000 bar with longer pulse durations. Overall, injection pressure plays the dominant role, whereas pulse duration and injection frequency determine the effective needle opening time and the sensitivity of the injection rate. Selecting the optimal parameter combinations for each frequency range ensures stable and repeatable fuel injection rates that meet operational requirements. Consequently, this study provides a scientific basis for subsequent research on machine-learning-based optimal control strategies for dual-fuel diesel injection systems, aiming to reduce emissions, enhance fuel economy, and improve engine performance.