Scholarlink

<p>Abrasive Water Jet Machining (AWJM) performance is strongly influenced by the condition of the focusing nozzle. As the focusing nozzle wears, the abrasive waterjet coherence is affected, leading to degradation in the machining quality. Real time prediction of focusing nozzle wear is necessary to support proactive maintenance actions. This study proposes a machine learning based edge computing framework for real time prediction of focusing nozzle wear. A dataset comprising 63 samples of vibration signal features, collected using a vibration sensor mounted at the cutting head for wear progression from 0 to 60&#xa0;h, was used to develop support vector classification, k-nearest neighbours, and logistic regression models. The vibration data were recorded at a high sampling rate of 50&#xa0;kHz, and the local processing was performed using a single board computer (SBC) with an edge computing architecture to avoid bandwidth related issues. From the acquired vibration signals, 22 time and frequency domain features were extracted, and these features were reduced into 7 principal components using Principal Component Analysis. To enhance prediction performance, model hyperparameters were optimized using a grid search based five fold cross validation procedure. Among the developed models, the support vector classifier achieved the highest prediction accuracy of 91% with its tuned hyperparameters: regularization value of 10, a kernel coefficient of 0.01, and linear kernel. The SBC implementation demonstrated an inference latency of 473.5 ms. This approach provides a practical solution for integrating real time wear prediction into AWJM systems, offering a scalable solution for intelligent maintenance in Industry 4.0 environments.</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

Edge computing based condition monitoring of abrasive waterjet focusing nozzle wear using machine learning

  • 期刊论文

  • 发表日期：

    2026年3月2日

查看全文
  • Sunilkumar Kothandaramaiah,
  • Kanthababu Mani

摘要

Abrasive Water Jet Machining (AWJM) performance is strongly influenced by the condition of the focusing nozzle. As the focusing nozzle wears, the abrasive waterjet coherence is affected, leading to degradation in the machining quality. Real time prediction of focusing nozzle wear is necessary to support proactive maintenance actions. This study proposes a machine learning based edge computing framework for real time prediction of focusing nozzle wear. A dataset comprising 63 samples of vibration signal features, collected using a vibration sensor mounted at the cutting head for wear progression from 0 to 60 h, was used to develop support vector classification, k-nearest neighbours, and logistic regression models. The vibration data were recorded at a high sampling rate of 50 kHz, and the local processing was performed using a single board computer (SBC) with an edge computing architecture to avoid bandwidth related issues. From the acquired vibration signals, 22 time and frequency domain features were extracted, and these features were reduced into 7 principal components using Principal Component Analysis. To enhance prediction performance, model hyperparameters were optimized using a grid search based five fold cross validation procedure. Among the developed models, the support vector classifier achieved the highest prediction accuracy of 91% with its tuned hyperparameters: regularization value of 10, a kernel coefficient of 0.01, and linear kernel. The SBC implementation demonstrated an inference latency of 473.5 ms. This approach provides a practical solution for integrating real time wear prediction into AWJM systems, offering a scalable solution for intelligent maintenance in Industry 4.0 environments.