T/R组件金丝键合焊点质量检测方法

Quality Inspection Method for Gold Wire Bonding Solder Joints in T/R Components

摘要:

针对发射/接收（T/R）组件中键合焊点的高效率准确检测需求，提出了一种结合外观特征和高维特征的无监督键合焊点质量检测方法。利用基于YOLOv8的键合焊点检测模型构建数据集，并采用霍夫圆变换（Hough circle transform，HCT）检测算法和视觉几何组（visual geometry group，VGG）卷积神经网络从焊点图像中提取了外观特征和高维特征。通过实验并结合实际情况设定了合理的阈值，构建了基于K-means算法的无监督键合焊点质量检测模型。测试结果表明，提出的检测方法取得了良好效果，有效解决了样本中严重的类别不平衡问题，为金丝键合焊点的质量控制提供了有效手段。

Abstract:

To meet the requirements of efficient and accurate detection for bonding solder joints in transmit/receive (T/R) components, an unsupervised bonding solder joints quality inspection method combining both appearance features and high-dimensional features was proposed. A dataset was constructed by a bonding solder joint inspection model based on YOLOv8. These appearance features and high-dimensional features were extracted from solder joint images using the Hough circle transform (HCT) algorithm and the visual geometry group (VGG) convolutional neural network (CNN). By setting reasonable thresholds based on experiments and practical considerations, an unsupervised bonding solder joints quality inspection model was established based on the K-means algorithm. The test results demonstrate that the proposed detection method achieves promising performance，effectively addressing the severe class imbalance issue in the dataset and providing an effective solution for quality control of gold wire bonding solder joints.