Scholarlink

<p>Capacitors in the Static Inverter (SIV), which are essential power conversion devices for railway vehicle operational stability, exhibit a high failure rate due to degradation and breakdown. This reveals the inefficiency of conventional time-based maintenance (TBM) and the limitations of existing diagnosis research conducted in restricted environments. To address these issues, this study proposes a data-driven AI model-based condition diagnosis system for railway vehicle SIV DC-Link capacitors. The proposed system estimates capacitance changes based on voltage, current, and temperature data collected during operation. Essential temperature compensation is applied, especially considering the temperature sensitivity of the externally installed SIV enclosure. For precise diagnosis, a high-sampling-frequency Data Sensing Module (DSM) is installed, and after deriving capacitance, data synchronized with the TCMS is secured for data preprocessing. The collected data are refined through median-based transformation, followed by temperature compensation, to analyze long-term degradation trends. A deep learning-based prediction model is employed to forecast future changes in capacitor state, and the performance of various regression algorithms, including Gradient Boosting, MLP, LSTM, and GPR, is compared. The proposed method was validated through field tests on AC/DC and DC railway vehicles, demonstrating its consistent applicability across different SIV systems and proving it to be a versatile capacitor diagnosis system.</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

Data-Driven AI Model for Railway Vehicle SIV Condition Diagnosis System

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • Hyo-Seok Oh,
  • Jae-Moon Kim

摘要

Capacitors in the Static Inverter (SIV), which are essential power conversion devices for railway vehicle operational stability, exhibit a high failure rate due to degradation and breakdown. This reveals the inefficiency of conventional time-based maintenance (TBM) and the limitations of existing diagnosis research conducted in restricted environments. To address these issues, this study proposes a data-driven AI model-based condition diagnosis system for railway vehicle SIV DC-Link capacitors. The proposed system estimates capacitance changes based on voltage, current, and temperature data collected during operation. Essential temperature compensation is applied, especially considering the temperature sensitivity of the externally installed SIV enclosure. For precise diagnosis, a high-sampling-frequency Data Sensing Module (DSM) is installed, and after deriving capacitance, data synchronized with the TCMS is secured for data preprocessing. The collected data are refined through median-based transformation, followed by temperature compensation, to analyze long-term degradation trends. A deep learning-based prediction model is employed to forecast future changes in capacitor state, and the performance of various regression algorithms, including Gradient Boosting, MLP, LSTM, and GPR, is compared. The proposed method was validated through field tests on AC/DC and DC railway vehicles, demonstrating its consistent applicability across different SIV systems and proving it to be a versatile capacitor diagnosis system.