A Dual-Attention Model for Non-Contact Multi-Fault PCB Diagnosis Based on Scaler Magnetic Field and Fault Relation Feature

The fault diagnosis method for printed circuit boards (PCBs) based on non-contact measurement data has significant application value in PCB fault diagnosis due to its non-destructive and efficient characteristics, especially true for radio frequency PCBs, where invisible areas are prominent, low-loss requirements are stringent, and frequency characteristics are complex. In this context, diagnosis based on electromagnetic field data shows unique advantages. However, most existing non-contact electromagnetic measurement instruments can only acquire scalar field data, limiting the performance of traditional diagnostic models. To address this issue, this paper proposes a method that integrates the scalar magnetic field spatial transformation, spatial distribution, and frequency distribution to construct a characteristic representation of the scalar magnetic field source described by the empirical cumulative distribution function. PCB fault relationship features, described using undirected graphs, are introduced. By leveraging convolutional graph neural networks relational graph parsing capabilities, these fault relationship features are combined with magnetic field source features, and a particular attention mechanism akin to boundary conditions is designed. Using the intrinsic relationship between PCB faults effectively constrains another enhanced Transformer training path, forming a dual-attention mechanism. This model reduces redundant focus on irrelevant information, improving the convergence speed and accuracy of the multi-fault diagnosis process. The innovation lies in breaking the traditional dependence of non-contact diagnostic models on singular measured data, substantially enhancing diagnostic comprehensive performance and practical engineering application value by fully integrating the inherent relationships between PCB faults.