1. Application of CNN-BP on Inconel-718 chip feature and the influence on tool life.
- Author
-
Chen, Shao-Hsien and Zhang, Ming-Jie
- Subjects
- *
CUTTING tools , *CONVOLUTIONAL neural networks , *NETWORKS on a chip , *PREDICTION models , *SIGNAL convolution , *SHEAR zones , *GEOMETRIC shapes , *BACK propagation - Abstract
The hard-to-cut material nickel-base Inconel-718 is used in large quantities in the aerospace industry and defense industry because the Inconel-718 has good mechanical strength at high temperatures, but the tool is likely to wear in the cutting processes. In this study, chip color and features were employed to predict tool lives because the chips are the closest to the cutting tool in the shear zone during the cutting processes, which can more authentically reflect the phenomenon of tool wear. The changes in chip color and geometric shape were extracted and digitized, and the prediction method employed a two-stage model. In Stage 1, a convolutional neural network (CNN) was used for chip condition classification and validation, and in Stage 2, back propagation neural network (BP) was employed to predict the tool wear. The result showed that when the Google-Net and ResNet50 network models were adopted for CNN chip feature recognition, with the confusion matrix of multi-classification for validation, the recognition precisions were 66.7% and 88.9%, respectively. The modeling and prediction were performed in the BP neural network; the chip chromaticity, chip thickness, and chip width were employed as input features, and the precision was enhanced by multi-fusion features. Finally, the mean absolute percentage error values for heavy cutting were 17.11%, medium cutting at 5.45%, and small cutting at 9.02%, indicating that the CNN-BP prediction model has good forecast accuracy, thus providing a new model prediction form for tool life. [ABSTRACT FROM AUTHOR]
- Published
- 2022
- Full Text
- View/download PDF