Feature Vector Identification and Prognostics of SAC305 PCBs for Varying G-Levels of Drop and Shock Loads

Abstract

This paper focuses on feature vector identification and prognostics of failure for SAC305 solder PCB's under varying conditions of shock loads. The test board is a multilayer FR4 of the JEDEC standard dimension with twelve packages arranged in a rectangular pattern. Strain signals and resistance measurements are acquired from different locations of the PCB for each drop. The strain signals are processed in the time and frequency domain to identify different feature vectors that can predict packages' failure. The varying drop load starts at 1500g and increases to 3000g, 5000g, and 7500g in a step increment manner until most of the board's packages fail. This experiment aims to understand the effectiveness of feature vectors in predicting failure at varying conditions of operating conditions and comparing the same with constant conditions. The strain signals that are acquired from various locations of the board are used to identify the failure of different packages on the board. Strain gauges are fixed on the backside of the PCB and to the package's corner positions, which are the most susceptible positions of failure. The feature vectors identified in the time and frequency domain are used to predict failure irrespective of changes in the operating conditions. The time-domain feature vector involves studying the characteristics of the peaks in the strain signal and the correlation of the corresponding peaks with an increase in the number of drops. Similarly, the frequency-domain feature vectors are obtained from statistical analysis on the frequency components, with and without the effect of the natural frequency of the board. The principal component analysis is used as the data reduction technique for both time and frequency domain analysis. The variation of the significant principal components is plotted to identify the patterns produced for before and after failure strain components. Four strain signals are used for a comparative study of feature vectors' behavior with multiple failures of packages on the board and the variation with different conditions of shock loads.