A novel eddy current sensor, featuring a triangular excitation coil and three rectangular receiving coils, is proposed for simultaneously detecting carbon fiber layer thickness and directions in the manufacturing process of carbon fiber reinforced polymer (CFRP) materials. It can disclose various carbon fiber layers and directions based on the signals obtained from the three receiving coils. However, classification of carbon fiber layers and directions proves challenging due to the complexity of signal features resulting from different combinations of carbon fiber layers and directions. To address this issue, we propose the use of a simplified multi-scale 1D-ResNet network. Comparative analysis reveals that our proposed network achieves higher train and validation accuracy compared to single-scale 1D-ResNet networks and different layers 1D-ResNet networks. Furthermore, the proposed simplified network not only achieves high classification accuracy with 100% and 99.7% for carbon fiber layers and directions, respectively, but also requires less training time compared to multi-scale 1D-ResNet networks. The novel eddy current sensor proposed in this study enables accurate recognition of carbon fiber layers and their directional orientation through the implementation of a simplified multi-scale 1D-ResNet network in the manufacturing process of CFRP materials. This innovation significantly enhances the quality of CFRP materials during manufacturing, ensuring their strength aligns with predetermined orientations.
Read full abstract