Scholarlink

<p>To enhance the fault detection performance of underwater thruster propeller winding and transmission abnormality faults, this paper presents an accurate diagnosis method for underwater thruster faults based on MTF and GAN-UTS. Initially, the MTF transformation technique is employed to convert one-dimensional propeller operation data into two-dimensional image information, thereby enhancing the intuitiveness of data analysis. Subsequently, the feature recognition capability of the GAN-UTS model is augmented through a transfer learning strategy. Finally, a Bidirectional Gated Recurrent Unit (BiGRU) Network integrated with a Spectral Attention Mechanism (SAM) is introduced at the model’s output to further improve fault diagnosis accuracy. To validate the practicality of this method, the study utilizes a dataset collected from the Jiaolong underwater vehicle and conducts a comparative analysis of this method against existing models such as CDAN, DDC, ACDANN, ADACL, BSP, DDTLN, and PGA. The experimental results demonstrate that, when addressing eight typical fault patterns, the proposed method outperforms other models in key performance indicators such as accuracy, precision, recall, and F1 score, thereby fully confirming the effectiveness and real-time performance of the employed approach.</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

Accurate diagnosis of underwater thruster faults based on MTF and GAN-UTS

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Xuyang Zhang,
  • Yu Zhang,
  • Linghui Peng,
  • Jining Peng

摘要

To enhance the fault detection performance of underwater thruster propeller winding and transmission abnormality faults, this paper presents an accurate diagnosis method for underwater thruster faults based on MTF and GAN-UTS. Initially, the MTF transformation technique is employed to convert one-dimensional propeller operation data into two-dimensional image information, thereby enhancing the intuitiveness of data analysis. Subsequently, the feature recognition capability of the GAN-UTS model is augmented through a transfer learning strategy. Finally, a Bidirectional Gated Recurrent Unit (BiGRU) Network integrated with a Spectral Attention Mechanism (SAM) is introduced at the model’s output to further improve fault diagnosis accuracy. To validate the practicality of this method, the study utilizes a dataset collected from the Jiaolong underwater vehicle and conducts a comparative analysis of this method against existing models such as CDAN, DDC, ACDANN, ADACL, BSP, DDTLN, and PGA. The experimental results demonstrate that, when addressing eight typical fault patterns, the proposed method outperforms other models in key performance indicators such as accuracy, precision, recall, and F1 score, thereby fully confirming the effectiveness and real-time performance of the employed approach.