With the increasing complexity of electronic circuit boards, traditional timing anomaly detection methods face significant challenges in high-dimensional data and complex signals, especially in real-time monitoring and fault warning systems. The focus of this work is the design of a Transformer-based temporal anomaly detection model that utilizes self-attention mechanism and Multilayer Perceptron to simultaneously learn long-term dependencies and local anomaly pattern while analyzing complex PCB temporal data. Inspired by the existing Transformer model architecture, we improve and propose the residual connections and multi-scale time window processing mechanism, which enables the model to better capture abnormal patterns and sudden faults in time series. Moreover, to increase the robustness of the model, we introduced a dynamic weighting mechanism to adaptively prioritise different time data attention whose purpose is to resist noise and abnormal fluctuations in time series data. Experiments demonstrate that the proposed model achieves a considerable enhancement in detection accuracy and robustness over traditional methods, especially for the data of abnormal circuit boards, the sensitivity and false alarm rate of the proposed model are significantly improved.