Scholarlink

Abrasive waterjet machining (AWJ) is used for cutting marbles, difficult to cut materials, super alloys, glass, plastics, etc. One of the major issues faced by the product in AWJ machining is the lower depth of smooth zone (DSZ). Also, due to inherent noise in the machine, DSZ also varies, giving product variability. Therefore, this study deals with in-process monitoring in AWJ machining to detect the depth of the smooth zone. Four input parameters are used in this work, giving 200 experiments. Each experiment gave a vibration signal, and five features were extracted from all the three domains of signal analysis. Empirical mode decomposition was used in time frequency domain. Fifteen features are fed as input to the deep neural network (DNN), random forest (RF), and decision tree (DT) by hyperparameter training. 80 models of DNN, 200 models of RF, and 48 models of DT are generated. Amongst all the models, RF gave a minimum value of mean absolute error of 0.8 and a maximum fitness coefficient of 0.96.

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

Application of Machine Learning and Empirical Mode Decomposition (EMD) in Prediction of Depth of Smooth Zone (DSZ) in AWJ Cutting by Vibration Signal

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Paramjit Thakur,
  • Yash Shinde,
  • Darshan Bhoir,
  • Devansh Bait

摘要

Abrasive waterjet machining (AWJ) is used for cutting marbles, difficult to cut materials, super alloys, glass, plastics, etc. One of the major issues faced by the product in AWJ machining is the lower depth of smooth zone (DSZ). Also, due to inherent noise in the machine, DSZ also varies, giving product variability. Therefore, this study deals with in-process monitoring in AWJ machining to detect the depth of the smooth zone. Four input parameters are used in this work, giving 200 experiments. Each experiment gave a vibration signal, and five features were extracted from all the three domains of signal analysis. Empirical mode decomposition was used in time frequency domain. Fifteen features are fed as input to the deep neural network (DNN), random forest (RF), and decision tree (DT) by hyperparameter training. 80 models of DNN, 200 models of RF, and 48 models of DT are generated. Amongst all the models, RF gave a minimum value of mean absolute error of 0.8 and a maximum fitness coefficient of 0.96.