Principal Component Analysis and Independent Component Analysis-Based Prognostic Health Monitoring of Electronic Assemblies Subjected to Simultaneous Temperature-Vibration Loads

This paper discusses methods for estimating different feature vectors from strain signals of an electronic assembly under combined temperature and vibration load. A vibrational load of 14 G acceleration-level with an ambient temperature of 55 °C is selected as the operating conditions for this experiment. Strain signals were measured at different time intervals during the vibration of the printed circuit board, and resistance values of the packages on the printed circuit board are monitored to identify the failure. The frequency response was measured by taking the fast Fourier transform of the signal and quantized by frequency quantization techniques. These techniques were able to identify the increase in the number of higher frequency components in the strain signal before failure with increase vibration time. The time-frequency response was also compared by employing different time–frequency analysis, joint time–frequency analysis, and statistical techniques such as principal component analysis (PCA), and independent component analysis (ICA). Statistical techniques like PCA and ICA were used to identify the different patterns of the original strain and filtered signals. These techniques discretely separated the before and after failure strain signals but were unable to predict the progression of failure in the packages. The instantaneous frequency of the strain signal displayed an interesting behavior, in which the variance of the PCA components of the instantaneous frequency had an increasing trend and reached a maximum value before continuously decreasing and reaching a lower value just before failure, indicating a progression of the before failure strain components.