As operating frequencies and clock speeds in processors have increased over the years, interconnects affect both the reliability and performance of entire electronic systems. Fault detection and diagnosis of the interconnects are crucial for prognostics and health management (PHM) of electronics. However, existing research works utilizing electrical signals as prognostic factors have limitations, such as the inability to distinguish the root cause of defects, which eventually requires additional destructive evaluation, and vulnerability to noise that results in a false alarm. Herein, we realize the non-destructive detection and diagnosis of defects in Cu interconnects, achieving early detection, high diagnostic accuracy, and noise robustness. To the best of our knowledge, this study first simultaneously analyzes the root cause and severity using electrical signal patterns. In this paper, we experimentally show that S-parameter patterns have the ability for fault diagnosis and they are effective input data for learning algorithms. Furthermore, we propose a novel severity rating ensemble learning (SREL) approach to enhance diagnostic accuracy and noise-robustness. Our method, with a maximum accuracy of 99.3%, outperforms conventional machine learning and multi-class convolutional neural networks (CNN) as additional noise levels increase.