Scholarlink

<p>In order to solve the problems of high data acquisition cost and low fault diagnosis rate in hydraulic system fault diagnosis, in this study, the hydraulic system of underwater vehicle water jet propulsion device is taken as the research object, a Hardware-in-the-Loop (HIL) method was employed to acquire a high-fidelity fault dataset. For accurate fault detection in the dataset, a hybrid fault diagnosis approach was proposed, which combines the Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM) network and Bayesian Optimization (BO) algorithm. The CNN-based approach is utilized for feature extraction, with subsequent fault classification performed by the BiLSTM model, while the Bo algorithm optimizes model hyperparameters to enhance accuracy. The experimental results indicate the Bo-CNN-BiLSTM model's diagnostic accuracy is 98.16%, which is 10.02%, 6.54%, and 3% higher than the accuracy of the BiLSTM model, CNN model, and CNN-BiLSTM model, respectively, and it demonstrating excellent recognition performance. This study provides methodological references for hydraulic systems fault diagnosis.</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 Identifying the Underwater Vehicle Water Jet Propulsion Device Hydraulic System Fault Based on Hardware-in-the-Loop and Bo-CNN-BiLSTM

  • 期刊论文

  • 发表日期：

    2026年4月17日

查看全文
  • Huawei Wang,
  • Linjia Nan,
  • Bo Wang,
  • Li Hu,
  • Xiaohui Luo

摘要

In order to solve the problems of high data acquisition cost and low fault diagnosis rate in hydraulic system fault diagnosis, in this study, the hydraulic system of underwater vehicle water jet propulsion device is taken as the research object, a Hardware-in-the-Loop (HIL) method was employed to acquire a high-fidelity fault dataset. For accurate fault detection in the dataset, a hybrid fault diagnosis approach was proposed, which combines the Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM) network and Bayesian Optimization (BO) algorithm. The CNN-based approach is utilized for feature extraction, with subsequent fault classification performed by the BiLSTM model, while the Bo algorithm optimizes model hyperparameters to enhance accuracy. The experimental results indicate the Bo-CNN-BiLSTM model's diagnostic accuracy is 98.16%, which is 10.02%, 6.54%, and 3% higher than the accuracy of the BiLSTM model, CNN model, and CNN-BiLSTM model, respectively, and it demonstrating excellent recognition performance. This study provides methodological references for hydraulic systems fault diagnosis.