Surface Roughness Prediction Based on CNN-BiTCN-Attention in End Milling

Surface roughness is a pivotal indicator of surface quality for machined components. It directly influences the performance and lifespan of manufactured products. Precise prediction of surface roughness is instrumental in refining production processes and curtailing costs. However, despite the use of identical processing parameters, the final surface roughness would be different. Thus, it challenges the effectiveness of traditional prediction models based solely on processing parameters. Current prevalent approaches for surface roughness prediction rely on handcrafted features, which require expert knowledge and considerable time investment. To address these challenges, we comprehensively consider the advantages of various deep learning methods and propose a novel end-to-end architecture. It synergistically integrates convolutional neural networks (CNN), bidirectional temporal convolutional networks (BiTCN), and attention mechanism, termed the CNN-BiTCN-Attention (CBTA) architecture. This architecture leverages CNN for automatic spatial feature extraction from signals, BiTCN to capture temporal dependencies, and the attention mechanism to focus on important features related to surface roughness. Experiments are conducted with popular deep learning methods on the public ACF dataset, which includes vibration, current, and force signals from the end milling process. The results demonstrate that the CBTA model outperforms other compared models. It achieves exceptional prediction performance with a mean absolute percentage error as low as 0.79% and an R2 as high as 99.81%. This validates the effectiveness and superiority of CBTA in end milling surface roughness prediction.

@article{Xiao2024,
  title     = {Surface Roughness Prediction Based on CNN-BiTCN-Attention in End Milling},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {12},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Guanhua Xiao and Hanqian Tu and Yunzhe Xu and Jiahao Shao and Dongming Xiang},
  doi       = {10.14569/IJACSA.2024.0151282},
  url       = {https://doi.org/10.14569/IJACSA.2024.0151282}
}