ML-Based Approach for PCB Anomaly and Leakage Detection
Matthieu Leflon, Mohamed Kheir, Sadok Ben Yahia · 2025
Machine Learning (ML) applications in our daily life have seen tremendous growth and they are closely tied to various Electromagnetic Interference and Compatibility (EMI/EMC) scenarios. This paper proposes a new ML-driven approach for detecting leakage and troubleshooting Printed Circuit Boards (PCBs) anomalies. The approach utilizes a simple Feedforward Neural Network (FNN) model that requires minimal training and a small dataset. This model is first trained on measurement points of 3D E-field data from an unslotted PCB. It then uses this training to detect the exact location of anomalies or E-field leakage within this PCB. This process achieves a high precision with a Root Mean Square Error (RMSE) of only 0.02 demonstrating an efficient detection accuracy.