Scholarlink

<p>Tool wear during machining has a significant impact on part quality and production efficiency. Traditional tool wear monitoring requires process interruptions, leading to inefficiency and underutilization of tool life. Real-time monitoring using sensor technologies and predictive modeling is crucial for improving machining efficiency and ensuring workpiece quality. This study investigates hybrid deep learning approaches for tool wear prediction, integrating Convolutional Neural Networks (CNN) with supervised learning models. The proposed method employs a CNN optimized using the Bayesian optimization (BOA) algorithm to extract image features. The BOA-CNN model is coupled with various supervised learning models, including Artificial Neural Networks (ANN), Support Vector Regression (SVR), Multi-layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), to incorporate temporal relationships between process variables and tool wear for enhancing the prediction efficiency. Experimental data collected from machining experiments conducted on IN718 superalloy by the milling process were used to train and test these models. The results indicate that the BOA-CNN model achieved the best performance for training data, with the lowest MAE (0.1038), MSE (0.0166), and RMSE (0.1289), with the highest R² value (0.9983). For testing data, the BOA-CNN-ANN model demonstrated superior performance compared to others, achieving the lowest MAE (0.3139) and RMSE (0.3946), as well as the highest R² (0.9787). While the CNN model alone has shown higher error metrics, the hybrid models demonstrate significant improvements. Further analysis of prediction intervals and error statistics reveals that the BOA-CNN model has the smallest confidence and prediction intervals during training, indicating superior reliability and accuracy. For testing data, the BOA-CNN-GRU model exhibited the highest consistency, with a small confidence interval (± 0.1212) and prediction interval (± 0.7634). The BOA-CNN-GRU model provided the most accurate and consistent tool wear predictions across both training and testing datasets. These models significantly outperform other deep learning models, making them the most suitable for real-time tool wear monitoring in machining processes. These findings highlight the potential of combining CNN with optimization techniques and supervised models to enhance predictive maintenance and minimize process disruptions in industrial settings.</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 approach concatenating CNN with supervised learning models for the tool wear prediction based on the analysis of tool wear image and temporal relationship among the cutting parameters

  • 期刊论文

  • 发表日期：

    2025年11月22日

查看全文
  • Abdul Arif,
  • Ponugoti Gangadhara Rao,
  • Kalapala Prasad

摘要

Tool wear during machining has a significant impact on part quality and production efficiency. Traditional tool wear monitoring requires process interruptions, leading to inefficiency and underutilization of tool life. Real-time monitoring using sensor technologies and predictive modeling is crucial for improving machining efficiency and ensuring workpiece quality. This study investigates hybrid deep learning approaches for tool wear prediction, integrating Convolutional Neural Networks (CNN) with supervised learning models. The proposed method employs a CNN optimized using the Bayesian optimization (BOA) algorithm to extract image features. The BOA-CNN model is coupled with various supervised learning models, including Artificial Neural Networks (ANN), Support Vector Regression (SVR), Multi-layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), to incorporate temporal relationships between process variables and tool wear for enhancing the prediction efficiency. Experimental data collected from machining experiments conducted on IN718 superalloy by the milling process were used to train and test these models. The results indicate that the BOA-CNN model achieved the best performance for training data, with the lowest MAE (0.1038), MSE (0.0166), and RMSE (0.1289), with the highest R² value (0.9983). For testing data, the BOA-CNN-ANN model demonstrated superior performance compared to others, achieving the lowest MAE (0.3139) and RMSE (0.3946), as well as the highest R² (0.9787). While the CNN model alone has shown higher error metrics, the hybrid models demonstrate significant improvements. Further analysis of prediction intervals and error statistics reveals that the BOA-CNN model has the smallest confidence and prediction intervals during training, indicating superior reliability and accuracy. For testing data, the BOA-CNN-GRU model exhibited the highest consistency, with a small confidence interval (± 0.1212) and prediction interval (± 0.7634). The BOA-CNN-GRU model provided the most accurate and consistent tool wear predictions across both training and testing datasets. These models significantly outperform other deep learning models, making them the most suitable for real-time tool wear monitoring in machining processes. These findings highlight the potential of combining CNN with optimization techniques and supervised models to enhance predictive maintenance and minimize process disruptions in industrial settings.