An informer network-based circuit boards fault detection method using infrared temperature series

As industrial demand for circuit board fault detection increases, infrared thermography has become a crucial non-invasive technique for the efficient identification of internal faults. However, existing methods exhibit limitations in feature extraction, local detail capture, and the modeling of correlations between chips and faults. To address these challenges, a comprehensive method that integrates a preprocessing stage and an enhanced Informer-based model, termed Informer-Fault-Net, is proposed. This method begins with preprocessing the long-term time-series heating data of components, which is collected by infrared cameras during power-on cycles. Subsequently, the processed data is fed into the Informer-Fault-Net model to identify faulty components on circuit boards. Within this network, a Statistic-SENet module is designed to pre-condition the input data by leveraging multiple statistical characteristics of component temperatures, and a channel attention mechanism is embedded within this module to strengthen the correlation between different chips and faults, thereby improving detection accuracy and robustness. Simultaneously, a Fully Convolutional Network (FCN) and an improved distillation mechanism are incorporated into the Informer encoder to enhance the model's capacity for local feature extraction and to reduce computational cost. A multi-scale feature fusion strategy is also employed to improve the model's ability to capture features across multiple scales. To validate the effectiveness of the proposed method, we designed and implemented an experimental hardware platform to collect a temperature time-series dataset from the components of circuit boards for fault detection. Finally, a series of experiments showed that the proposed method achieved an accuracy of 0.990.