Crack Diagnosis of Surface-Mount Capacitors Using AI Classification Models with Multi-Parameter Impedance Spectra

Surface-mounted devices (SMDs) are essential components that enable the miniaturization and enhanced performance in electronic products, significantly impacting both circuit performance and reliability. In this study, we propose a non-destructive evaluation method for cracks in SMD capacitors using the artificial intelligence of impedance spectra. To achieve this, cracks were induced in 132 specimens through incremental displacement using a shear module of a bond tester. At each crack level, frequency-domain spectra were acquired for 14 parameters using an impedance analyzer. Meaningful changes in parameter patterns corresponding to each crack stage were observed, confirming impedance spectroscopy as an effective tool for crack assessment. Through data augmentation, we generated 87,800 datasets representing various crack stages, which were used to train AI models that output crack stages from input impedance spectra. Based on this dataset, six AI models, ConvNeXt, LSTM, Transformer, Logistic Regression, SVM, and Random Forest, were developed to classify crack severity into nine stages. Model-wise, the Random Forest classifier consistently outperformed the other approaches. When trained with single parameters, it achieved its best performance using the dissipation factor, reaching 98.5% accuracy. Furthermore, when the dissipation factor was combined with any of the remaining impedance parameters, the Random Forest model achieved perfect diagnostic performance (100%) across all combinations, highlighting both its robustness and its suitability for multi-parameter learning. These results provide practical guidance for selecting effective parameters and model architectures for impedance spectrum-based crack diagnostics.