Core Concepts
Tensor Convolutional Neural Networks (T-CNNs) can achieve the same performance as classical Convolutional Neural Networks (CNNs) for defect detection in manufacturing, while using up to 15 times fewer parameters and training 4-19% faster.
Abstract
The authors introduce a Tensor Convolutional Neural Network (T-CNN) and examine its performance on a real defect detection application in the manufacturing of ultrasonic sensors.
Key highlights:
- The T-CNN operates on a reduced model parameter space to substantially improve the training speed and performance of an equivalent CNN model without sacrificing accuracy.
- The T-CNN is able to reach the same performance as classical CNNs as measured by quality metrics, with up to fifteen times fewer parameters and 4% to 19% faster training times.
- The T-CNN greatly outperforms the results of traditional human visual inspection, providing value in a current real application in manufacturing.
- There is a trade-off between model performance and computational efficiency as the rank of the tensor convolutional layers is decreased. However, the authors identify an optimal rank configuration that maintains high quality metrics while offering significant reductions in parameters and training time.
- Integrating the T-CNN in quality control systems can free up human resources for more cognitive tasks, leading to increased efficiency and productivity in manufacturing.
Stats
The dataset contains a total of 11,728 labeled images of ultrasonic sensor components, with 9 possible types of defects.
The T-CNN with rank configuration (32, 32, 3, 3) has 4.6 times fewer parameters than the CNN.
The T-CNN with rank configuration (32, 32, 3, 3) has a 16% faster training time compared to the CNN.
Quotes
"Tensor Convolutional Neural Networks (T-CNNs) can achieve the same performance as classical Convolutional Neural Networks (CNNs) for defect detection in manufacturing, while using up to 15 times fewer parameters and training 4-19% faster."
"The T-CNN greatly outperforms the results of traditional human visual inspection, providing value in a current real application in manufacturing."