Scholarlink

<p>Tool life critically influences the efficiency and precision of automated manufacturing, directly impacting workpiece quality, dimensional accuracy, and operational costs. To address the challenge of tool life prediction with limited data, this study introduces a&#xa0;hybrid deep learning model optimized for milling IN718 superalloy. Vibration signals from cutting tools were processed using Gabor Wavelet Transform (GWT) for feature extraction, followed by Kernel based principal component analysis (KPCA) for dimensionality reduction. A&#xa0;Bayesian Optimization Algorithm (BOA) fine-tuned the model’s hyperparameters, achieving an optimal configuration with a&#xa0;learning rate of 8.22 × 10⁻<sup>3</sup>, 220 hidden units, and 2&#xa0;hidden layers. The proposed GWT+KPCA+BOA-BiLSTM hybrid model achieved an average RMSE of 0.006744 and a&#xa0;prediction accuracy (R<sup>2</sup>) of 90.11% using hybrid features, compared to an RMSE of 0.009182 and an R<sup>2</sup> of 85.89% using automatic features, clearly demonstrating its superior performance over the automatic feature-based model. This approach enables reliable tool wear monitoring even with sparse datasets, enhancing predictive maintenance in smart manufacturing.</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

Hybrid deep learning for cutting tool life monitoring and prediction in IN718 milling using limited data

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Mulpur Sarat Babu

摘要

Tool life critically influences the efficiency and precision of automated manufacturing, directly impacting workpiece quality, dimensional accuracy, and operational costs. To address the challenge of tool life prediction with limited data, this study introduces a hybrid deep learning model optimized for milling IN718 superalloy. Vibration signals from cutting tools were processed using Gabor Wavelet Transform (GWT) for feature extraction, followed by Kernel based principal component analysis (KPCA) for dimensionality reduction. A Bayesian Optimization Algorithm (BOA) fine-tuned the model’s hyperparameters, achieving an optimal configuration with a learning rate of 8.22 × 10⁻3, 220 hidden units, and 2 hidden layers. The proposed GWT+KPCA+BOA-BiLSTM hybrid model achieved an average RMSE of 0.006744 and a prediction accuracy (R2) of 90.11% using hybrid features, compared to an RMSE of 0.009182 and an R2 of 85.89% using automatic features, clearly demonstrating its superior performance over the automatic feature-based model. This approach enables reliable tool wear monitoring even with sparse datasets, enhancing predictive maintenance in smart manufacturing.