Scholarlink

<p>Defect detection in castings is essential for ensuring structural reliability in quality-critical industries. Traditional deep learning approaches face challenges such as high computational cost, susceptibility to noise, and overfitting when trained on limited datasets. To address these issues, we propose a hybrid quantum–classical framework for automated casting defect detection using quantum-inspired neural networks (QNNs). Here, “quantum-inspired” refers to algorithms based on principles of quantum walks and variational quantum circuits, implemented on classical simulation hardware. The model combines classical preprocessing with quantum variational layers to classify defects such as porosity, shrinkage, and micro-cracks from X-ray images. Experiments on the GDXray Castings dataset show that the eight-qubit QNN achieved 93.8% accuracy, 92.7% precision, 94.5% recall, and an F1-score of 93.6%, surpassing a baseline convolutional neural network. This work provides the first reported use of QNNs for casting inspection, offering a promising pathway toward robust and scalable non-destructive testing solutions 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

Quantum neural networks for casting defect detection: a hybrid intelligence framework for smart manufacturing

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Nabhan Yousef,
  • Chandrasinh Parmar,
  • Amit Sata,
  • Abhilash Edacherian

摘要

Defect detection in castings is essential for ensuring structural reliability in quality-critical industries. Traditional deep learning approaches face challenges such as high computational cost, susceptibility to noise, and overfitting when trained on limited datasets. To address these issues, we propose a hybrid quantum–classical framework for automated casting defect detection using quantum-inspired neural networks (QNNs). Here, “quantum-inspired” refers to algorithms based on principles of quantum walks and variational quantum circuits, implemented on classical simulation hardware. The model combines classical preprocessing with quantum variational layers to classify defects such as porosity, shrinkage, and micro-cracks from X-ray images. Experiments on the GDXray Castings dataset show that the eight-qubit QNN achieved 93.8% accuracy, 92.7% precision, 94.5% recall, and an F1-score of 93.6%, surpassing a baseline convolutional neural network. This work provides the first reported use of QNNs for casting inspection, offering a promising pathway toward robust and scalable non-destructive testing solutions in smart manufacturing.