Scholarlink

<p>This paper addresses the critical need for real-time health monitoring in flyback converters, key components in renewable energy systems. Conventional methods often require additional sensors, increasing cost and complexity. To overcome this, this paper proposes a novel real-time artificial neural network (ANN) for online parameter estimation of a digital twin, enabling sensorless health indication. The digital twin is derived from the state-space averaged (SSA) model of the converter, and its dynamics are solved using the 4th-order Runge-Kutta method. A multilayer perceptron (MLP) network with backpropagation is employed as the ANN, uniquely trained online during simulation to estimate critical component parameters—including capacitance, semiconductor on-state resistance, diode forward voltage, and transformer magnetization inductance—by minimizing the error between the simulated converter and digital twin outputs. The proposed method was rigorously tested in a MATLAB/Simulink environment under various scenarios, such as component degradation and changes in input voltage and duty cycle. Results demonstrate the ANN’s efficacy, achieving accurate parameter estimation with an average error of less than 1% for capacitance degradation. This method offers a significant advantage by eliminating the need for extensive offline training data and additional hardware, presenting a more adaptive and efficient solution for converter health monitoring.</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

A real-time ANN for estimation of digital twin parameters used in health indication of flyback converter

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月16日

查看全文
  • Ehsan Rahmanian Koushkaki,
  • Amirhosein Rajaei

摘要

This paper addresses the critical need for real-time health monitoring in flyback converters, key components in renewable energy systems. Conventional methods often require additional sensors, increasing cost and complexity. To overcome this, this paper proposes a novel real-time artificial neural network (ANN) for online parameter estimation of a digital twin, enabling sensorless health indication. The digital twin is derived from the state-space averaged (SSA) model of the converter, and its dynamics are solved using the 4th-order Runge-Kutta method. A multilayer perceptron (MLP) network with backpropagation is employed as the ANN, uniquely trained online during simulation to estimate critical component parameters—including capacitance, semiconductor on-state resistance, diode forward voltage, and transformer magnetization inductance—by minimizing the error between the simulated converter and digital twin outputs. The proposed method was rigorously tested in a MATLAB/Simulink environment under various scenarios, such as component degradation and changes in input voltage and duty cycle. Results demonstrate the ANN’s efficacy, achieving accurate parameter estimation with an average error of less than 1% for capacitance degradation. This method offers a significant advantage by eliminating the need for extensive offline training data and additional hardware, presenting a more adaptive and efficient solution for converter health monitoring.