Scholarlink

<p>Manufacturing sector is estimated to generate approximately 2000&#xa0;PB of metadata annually in the near term from both industry and academic activities. Despite this vast data production, a persistent challenge remains in digital manufacturing is that most of the collected manufacturing data are ‘information absent’, primarily due to limitations on data collection methods and data privacy concerns. The information absent data typically lacks one or two critical pieces of information, such as geometric or dimensional features of manufactured products, either for a single data point or specific datasets. The absence of such essential information further deteriorates the long-standing challenge on the significant shortage of labelled manufacturing data. To process the information absent data, the evolutionary binary (EB) algorithm was proposed following thermo-mechanical principles. In a case study, this algorithm enabled the recognition of essential geometric features from an information absent dataset of hot stamping process by labelling the origins of each data point. Serving as a highly flexible framework, the EB algorithm enables the incorporation of a variety of thermo-mechanical filters (TMFs) developed based on data analysis. Results demonstrate that, with the integration of TMFs, the EB algorithm achieved an overall accuracy of nearly 95% with an average computational speed approximately 100 times faster than that of classic machine learning algorithms, despite using extremely sparse labelled data.</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

Developing thermo-mechanical filters (TMFs) for recognising information absent hot stamping data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月10日

查看全文
  • Heli Liu,
  • Xiaochuan Liu,
  • Denis J. Politis,
  • Xiao Yang,
  • Yang Zheng,
  • Liliang Wang

摘要

Manufacturing sector is estimated to generate approximately 2000 PB of metadata annually in the near term from both industry and academic activities. Despite this vast data production, a persistent challenge remains in digital manufacturing is that most of the collected manufacturing data are ‘information absent’, primarily due to limitations on data collection methods and data privacy concerns. The information absent data typically lacks one or two critical pieces of information, such as geometric or dimensional features of manufactured products, either for a single data point or specific datasets. The absence of such essential information further deteriorates the long-standing challenge on the significant shortage of labelled manufacturing data. To process the information absent data, the evolutionary binary (EB) algorithm was proposed following thermo-mechanical principles. In a case study, this algorithm enabled the recognition of essential geometric features from an information absent dataset of hot stamping process by labelling the origins of each data point. Serving as a highly flexible framework, the EB algorithm enables the incorporation of a variety of thermo-mechanical filters (TMFs) developed based on data analysis. Results demonstrate that, with the integration of TMFs, the EB algorithm achieved an overall accuracy of nearly 95% with an average computational speed approximately 100 times faster than that of classic machine learning algorithms, despite using extremely sparse labelled data.