Scholarlink

<p>To address the issue that sensors in the Flush air data sensing (FADS) system are vulnerable to faults, which impairs the accuracy of parameter calculation, this study introduces a novel method combining ensemble learning with a CNN-Transformer. The ensemble learning model is employed to identify whether the FADS pressure measurements are faulty. Based on the detection results, a CNN Transformer architecture with a dynamically adjustable input structure is designed to adaptively select valid sensor data as inputs. Experimental results demonstrate that the fault detection module achieves an accuracy of 0.9875 in single fault scenarios and 0.9556 in complex multi fault scenarios, outperforming comparison methods. On this basis, the fault-tolerant network provides reliable air data estimation during concurrent sensor failures. Moreover, in the extreme case of 6 faults, the prediction errors for both the Angle of attack and Angle of sideslip are less than 0.5 degrees, meeting flight control requirements. Numerical simulations verify that the algorithm achieves accurate fault identification in multi-fault scenarios and high-precision fault-tolerant solution of key air data parameters such as Mach number, angle of attack, and static pressure.</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

An approach to fault detection and dynamic fault tolerance for flush air data sensing system with multiple faults

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月2日

查看全文
  • Dibo Xiao,
  • Jie Wang,
  • Junjie Wang

摘要

To address the issue that sensors in the Flush air data sensing (FADS) system are vulnerable to faults, which impairs the accuracy of parameter calculation, this study introduces a novel method combining ensemble learning with a CNN-Transformer. The ensemble learning model is employed to identify whether the FADS pressure measurements are faulty. Based on the detection results, a CNN Transformer architecture with a dynamically adjustable input structure is designed to adaptively select valid sensor data as inputs. Experimental results demonstrate that the fault detection module achieves an accuracy of 0.9875 in single fault scenarios and 0.9556 in complex multi fault scenarios, outperforming comparison methods. On this basis, the fault-tolerant network provides reliable air data estimation during concurrent sensor failures. Moreover, in the extreme case of 6 faults, the prediction errors for both the Angle of attack and Angle of sideslip are less than 0.5 degrees, meeting flight control requirements. Numerical simulations verify that the algorithm achieves accurate fault identification in multi-fault scenarios and high-precision fault-tolerant solution of key air data parameters such as Mach number, angle of attack, and static pressure.