基于Transformer的DC/DC板级验证状态识别

DC/DC board-level verification status recognition based on Transformer

为满足航天产品的高精度、高可靠性需求,实现元器件自主可控、芯片国产化及应用适应性验证十分必要,设计一种基于FPGA的国产DC/DC板级综合测试平台。在长时间的热学环境适应性板级验证项目中,为实现DC/DC器件应用板卡工作状态的实时监测,提出一种基于Transformer的智能识别算法。分别使用空载、负载电流3 A、负载电流5 A、高输入电压、低输入电压、短路状态下的DC DC输出序列,输入到Transformer模型中并利用注意力机制提取各序列的全局注意力特征,并对深度学习模型进行训练。实验结果表明,对于此6种工作状态数据集,Transformer模型识别的准确率为99.2%,具备良好的分类和监测性能,具有一定的工程应用价值。

To meet the high-precision and high-reliability requirements of aerospace products,and achieve self-reliance in components and chip localization,as well as application adaptability verification,it is necessary to design a domestic DC/DC board-level comprehensive testing platform based on FPGA.In long-term thermal environmental adaptability board-level verification projects,to achieve real-time monitoring of the working status of DC/DC device application boards,a smart recognition algorithm based on Transformer is proposed.The deep learning model is trained using these features.The experiment results show that for this 6-state dataset,the recognition accuracy of the Transformer model is 99.2%,which has good classification and monitoring performance and has certain engineering application value.