Scholarlink

<p>This study presents for the first time a framework that is entirely based on data-driven methods for constructing the load spectrum of a CNC lathe. Firstly, the CNN-LSTM-Transformer (CLT) model is utilized to accurately reconstruct a more representative full life cycle load history from sparse short-term data. Subsequently, the CNN-Normalizing Flow (CNF) model is employed to directly learn the joint probability density of the load mean amplitude matrix through an invertible generative flow. This enables accurate extrapolation of the two-dimensional spectrum in a non-parametric manner, effectively circumventing the limitations of traditional edge distribution fitting. The final experimental results show that this approach can reduce the RMSE by 25.63 %, the MAPE by 20.93 %, while increasing the R<sup>2</sup> by 1.92 %. This framework provides a data-driven reliable solution for load spectrum construction and reliability assessment in engineering applications.</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

Load extrapolation method for CNC lathe based on CLT-CNF model

  • 期刊论文

  • 发表日期：

    2026年8月5日

查看全文
  • Xingtao Xu,
  • Chenchen Wu,
  • Wenzhi Cao,
  • Linjie Zhang,
  • Yan Liu,
  • Jialong He

摘要

This study presents for the first time a framework that is entirely based on data-driven methods for constructing the load spectrum of a CNC lathe. Firstly, the CNN-LSTM-Transformer (CLT) model is utilized to accurately reconstruct a more representative full life cycle load history from sparse short-term data. Subsequently, the CNN-Normalizing Flow (CNF) model is employed to directly learn the joint probability density of the load mean amplitude matrix through an invertible generative flow. This enables accurate extrapolation of the two-dimensional spectrum in a non-parametric manner, effectively circumventing the limitations of traditional edge distribution fitting. The final experimental results show that this approach can reduce the RMSE by 25.63 %, the MAPE by 20.93 %, while increasing the R2 by 1.92 %. This framework provides a data-driven reliable solution for load spectrum construction and reliability assessment in engineering applications.