Scholarlink

This study presents a machine learning (ML)-assisted approach to optimize a corona-based diesel particulate matter (PM) removal system. Building on an established non-thermal plasma model, simulation data integrated with ML to improve electrostatic precipitator (ESP) design under power constraints. A MATLAB-based simulator was used to model 3,996 configurations, considering electrical, geometric, and engine parameters. Two multi-layer perceptron regressors were trained to predict efficiency and power consumption. Using the models, 5,000 new configurations were generated, and those exceeding 500 W power usage were discarded. The top 5 remaining setups achieved up to 97% predicted efficiency with less than 500 W power use. Optimal designs featured small discharge wire radii, medium collection cylinder sizes, and extended charging zones. This AI-driven method reduces design iteration time, highlights optimal trade-offs, and supports real-time decision-making for advanced PM removal systems. Study showed that high efficiency designs are possible even under strict power limits and that electrode geometry plays a major role.

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

Optimizing a Diesel Aftertreatment System Using Machine Learning and Simulation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Aleksandr Šabanovič,
  • Jonas Matijošius,
  • Artūras Kilikevičius,
  • Aleksandras Chlebnikovas,
  • Tadas Žvirblis

摘要

This study presents a machine learning (ML)-assisted approach to optimize a corona-based diesel particulate matter (PM) removal system. Building on an established non-thermal plasma model, simulation data integrated with ML to improve electrostatic precipitator (ESP) design under power constraints. A MATLAB-based simulator was used to model 3,996 configurations, considering electrical, geometric, and engine parameters. Two multi-layer perceptron regressors were trained to predict efficiency and power consumption. Using the models, 5,000 new configurations were generated, and those exceeding 500 W power usage were discarded. The top 5 remaining setups achieved up to 97% predicted efficiency with less than 500 W power use. Optimal designs featured small discharge wire radii, medium collection cylinder sizes, and extended charging zones. This AI-driven method reduces design iteration time, highlights optimal trade-offs, and supports real-time decision-making for advanced PM removal systems. Study showed that high efficiency designs are possible even under strict power limits and that electrode geometry plays a major role.