Scholarlink

<p>Large-scale mechanical fault detection plays a critical role in industrial production, and neural networks have been widely adopted for this task. However, the performance of neural networks heavily depends on the optimization of their architecture and parameters. Traditional optimization methods such as gradient descent and particle swarm optimization can improve model performance to some extent, but they typically require adjusting all weights, which not only increases computational complexity but also often leads to redundant network structures. To address these challenges, this paper proposes a cooperative neuroevolution algorithm for constructing lightweight neural networks tailored for fault diagnosis tasks. The algorithm introduces two co-evolving populations-one dedicated to learning the critical weights of the neural network, and the other focused on optimizing the precise weight values. This approach enables the discovery of a sparsely connected, lightweight neural network. By decoupling and cooperatively optimizing network structure and weights, the algorithm can more efficiently explore the search space to identify models that are both compact and high-performing. To validate the effectiveness of the proposed algorithm, extensive experiments are conducted on multiple industrial datasets involving large-scale mechanical fault detection. The results demonstrate that the algorithm maintains high diagnostic accuracy while significantly reducing the number of model parameters. Compared to conventional evolutionary algorithms, it exhibits superior model compression capability and generalization performance.</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 Lightweight Mechanical Fault Diagnosis Network Driven by Cooperative Neuroevolution Algorithm

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Haifeng Yue,
  • Jue Zhang,
  • Yize Wang,
  • Xiaojian Tian,
  • Bairong Zhang,
  • Zhennan Zhang,
  • Shuai Shao

摘要

Large-scale mechanical fault detection plays a critical role in industrial production, and neural networks have been widely adopted for this task. However, the performance of neural networks heavily depends on the optimization of their architecture and parameters. Traditional optimization methods such as gradient descent and particle swarm optimization can improve model performance to some extent, but they typically require adjusting all weights, which not only increases computational complexity but also often leads to redundant network structures. To address these challenges, this paper proposes a cooperative neuroevolution algorithm for constructing lightweight neural networks tailored for fault diagnosis tasks. The algorithm introduces two co-evolving populations-one dedicated to learning the critical weights of the neural network, and the other focused on optimizing the precise weight values. This approach enables the discovery of a sparsely connected, lightweight neural network. By decoupling and cooperatively optimizing network structure and weights, the algorithm can more efficiently explore the search space to identify models that are both compact and high-performing. To validate the effectiveness of the proposed algorithm, extensive experiments are conducted on multiple industrial datasets involving large-scale mechanical fault detection. The results demonstrate that the algorithm maintains high diagnostic accuracy while significantly reducing the number of model parameters. Compared to conventional evolutionary algorithms, it exhibits superior model compression capability and generalization performance.