Optimizing Metal Cutting Empirical tests & AI-enhanced tool wear detection
(2024) MMTM05 20241Production and Materials Engineering
- Abstract
- Detection of tool wear is critical in optimizing tool change strategies and cutting parameter selection to influence productivity, cost, and quality in metal cutting. Nonetheless, current practices for tool wear detection majorly depend on manual, inconsistent, and time-consuming visual inspection by operators.
Tool wear detection has the potential to greatly benefit from recent advances in artificial intelligence, computer vision, and deep learning. This thesis investigates the use of deep learning and image classification techniques to detect tool wear in metal cutting. The research involves performing empirical machining tests to create a diverse dataset of tool wear images under various cutting conditions, tool geometries, and... (More) - Detection of tool wear is critical in optimizing tool change strategies and cutting parameter selection to influence productivity, cost, and quality in metal cutting. Nonetheless, current practices for tool wear detection majorly depend on manual, inconsistent, and time-consuming visual inspection by operators.
Tool wear detection has the potential to greatly benefit from recent advances in artificial intelligence, computer vision, and deep learning. This thesis investigates the use of deep learning and image classification techniques to detect tool wear in metal cutting. The research involves performing empirical machining tests to create a diverse dataset of tool wear images under various cutting conditions, tool geometries, and workpiece materials. The images are annotated with the corresponding wear metrics such as flank wear, crater wear, and wear morphology.
Using the PyTorch deep learning framework in Python, we create and train a Convolutional Neural Network (CNN) on the tool wear image dataset. Transfer learning is used to fine-tune pretrained CNN architectures for the tool wear classification task. To make the model robust and improve generalization, data augmentation and cross-validation are implemented. The performance of the CNN model is evaluated on unseen test images and benchmarked against traditional computer vision methods.
In addition, the thesis investigates the relationship between the extracted CNN features, the cutting parameters, and the wear values measured. To interpret the CNN detections, the wear images are analyzed using Gradient-weighted Class Activation Mapping (Grad-CAM) and the regions of interest are highlighted. The proposed deep learning-based tool wear detection approach offers encouraging results compared to manual examination in terms of accuracy, reliability, and efficiency. The development demonstrates the potential for integration with actual tool condition monitoring and cutting process optimization systems. Hence, this investigation provides groundbreaking insights into the field of smart manufacturing and Industry 4.0. It offers clear recommendations on AI-driven manufacturing to strengthen efficiency. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9158088
- author
- Sayegh, Anas LU
- supervisor
-
- Volodymyr Bushlya LU
- Petra Maier LU
- organization
- course
- MMTM05 20241
- year
- 2024
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Tool wear, metal cutting, deep learning, convolutional neural networks, image classification, PyTorch, smart manufacturing.
- report number
- LUTMDN/(TMMV-5365)/1-59/2024
- language
- English
- id
- 9158088
- date added to LUP
- 2024-06-04 18:33:55
- date last changed
- 2024-06-05 11:11:54
@misc{9158088,
abstract = {{Detection of tool wear is critical in optimizing tool change strategies and cutting parameter selection to influence productivity, cost, and quality in metal cutting. Nonetheless, current practices for tool wear detection majorly depend on manual, inconsistent, and time-consuming visual inspection by operators.
Tool wear detection has the potential to greatly benefit from recent advances in artificial intelligence, computer vision, and deep learning. This thesis investigates the use of deep learning and image classification techniques to detect tool wear in metal cutting. The research involves performing empirical machining tests to create a diverse dataset of tool wear images under various cutting conditions, tool geometries, and workpiece materials. The images are annotated with the corresponding wear metrics such as flank wear, crater wear, and wear morphology.
Using the PyTorch deep learning framework in Python, we create and train a Convolutional Neural Network (CNN) on the tool wear image dataset. Transfer learning is used to fine-tune pretrained CNN architectures for the tool wear classification task. To make the model robust and improve generalization, data augmentation and cross-validation are implemented. The performance of the CNN model is evaluated on unseen test images and benchmarked against traditional computer vision methods.
In addition, the thesis investigates the relationship between the extracted CNN features, the cutting parameters, and the wear values measured. To interpret the CNN detections, the wear images are analyzed using Gradient-weighted Class Activation Mapping (Grad-CAM) and the regions of interest are highlighted. The proposed deep learning-based tool wear detection approach offers encouraging results compared to manual examination in terms of accuracy, reliability, and efficiency. The development demonstrates the potential for integration with actual tool condition monitoring and cutting process optimization systems. Hence, this investigation provides groundbreaking insights into the field of smart manufacturing and Industry 4.0. It offers clear recommendations on AI-driven manufacturing to strengthen efficiency.}},
author = {{Sayegh, Anas}},
language = {{eng}},
note = {{Student Paper}},
title = {{Optimizing Metal Cutting Empirical tests & AI-enhanced tool wear detection}},
year = {{2024}},
}