An intelligent detection method for assembly based on multi-model cascade

In the assembly process of the anti-loosening steel wire of the aero-engine variable stator vane (VSV) adjusting rod, it is still necessary to manually detect the correctness of the assembly, which is inefficient and error-prone. To replace manual detection, an intelligent detection method based on a multi-model cascade is proposed. The method contains two parts: a detection module and a classification module. Firstly, on the detection module, we propose to improve YOLOv5s by mixing depth-separable convolution with different-size convolutional kernels and lightweight decoupling head, and the improved YOLOv5s achieves an average accuracy of 97.9% on the test set, which improves by 3.4% and 1.5% compared to YOLOv5s and YOLOv8s, respectively. Secondly, the ConvNeXt classification head is improved by using 7*7 deep convolution instead of global average pooling on the classification module, and the performance is improved to reach 97.5% and 95.4% accuracy on the two datasets, respectively. Finally, the detection method is utilized to validate the image dataset collected by the on-site assembly workshop, and the results show that the average accuracy of this paper's method reaches 92.7%, which further validates the reliability of this paper's intelligent assembly detection method.